235

SOME EXPERIENCE WITH VENTILATION CONTROLBASED ON TRAFFIC DATA†

Eddy Jacques, Université Catholique de Louvain & Louis Possoz, Mitec sa, Belgium

ABSTRACT

More and more tunnels are equipped to measure traffic flows. Unfortunately, this equipmentoperates often in a stand-alone mode, the first concern being to display and store the measuredtraffic data. Nevertheless, one could gain more out of this information by using it directly orindirectly for controlling the ventilation.

So far, ventilation control, if any, has been based on pollutant concentration measurements.However, this technique shows a few weaknesses :• For certain pollutants, such as NO2, there is a real difficulty to measure reliably their

concentrations. Yet, it looks like that this component will play a more important role thanCO in the near future. In the meantime, an estimation of these concentrations by anexperimental model could yield better results;

• Sometimes, and for understandable reasons, the measurement devices are not installed attheir most optimal location. This shortcoming could cause the measured value not beingalways representative of the average one;

• In case of variable traffic conditions, the variations are perceived much faster on a trafficdiagram than they would be on a pollution diagram. This fact allows the control system toreact more rapidly either to the increasing pollution, even with the possibility to anticipateit, or to the excessive air quantities in the opposite case.

In this paper, with the example of the Belliard tunnel at hand, the advantages of using trafficdata instead of pollution data to control the ventilation will become obvious. The exampleshows not only a gain in terms of energy saving, but also in terms of a shorter response timeof the system.

After a brief description of the tunnel, its lay-out and installed equipment, the controlalgorithms are described. They are mainly characterised by a centralised, discrete and twolevel control mode : density and speed characteristics of the traffic are used as drivingvariables; the averaged pollutant concentrations as supervising variables.

† Paper published in "Tunnel Control & Communication", A.Vardy Edt., 2nd Int. Conference, Amsterdam, 1997

236

1 INTRODUCTION

Traffic is undoubtedly one of the most important sources of environmental nuisance in our citiesand densely populated areas. It does not only have a damaging impact on health through noiseand air pollution, but also indirectly debases quality of life by disturbing the eco-system,destroying our urbanistic and cultural inheritance, and so on.

Despite all efforts and all kinds of recommendations to reduce both traffic flow and its resultingnuisances, it may be expected for the next few decades that traffic remains an important sourceof environmental aggression. Therefore a lot of money has been invested in building roadtunnels which is a manner of responding to this unavoidable problem. However, their number issteadily growing. These tunnels will ultimately be interconnected, so creating a realunderground network, the complexity of which, from the ventilation point of view, should not beunderestimated. The Belliard tunnel in Brussels is one representative example of this kind ofcomplex tunnels.

The concern for a healthy and safe environment is now extended to the inside of the tunnels,with an even bigger challenge. One has not only to maintain a high standard of air quality, butalso good visibility, and an acceptable noise level. Well studied and carefully designed tunnelinfrastructures should reduce all potential hazards. Equally important, is the efficiency andreliability of its management and control system.

Since tunnel users are ever more demanding, especially when their health and safety areconcerned, there is an undeniable need to install sophisticated and expensive equipment such ashigh-sensitive pollution sensors, reliable fire detection systems, traffic sensing devices and videocameras.Unfortunately, most of this equipment, sometimes installed at different life stages of the tunnel,operates often in a stand-alone mode, the first concern being only the display and the print-out ofthe measured data.

However, with increasing complexity of tunnels, the supervision of all this installed equipmentand devices becomes undoubtedly more difficult. Therefore, one needs to introduce computertechnology, which at one go allows us to gain a lot more from the measurements and alarms thana mere display of status information. Indeed, connecting these devices to a central supervisionunit, which could be an expert system, has the advantage to concentrate and process togethermeasured information of different nature and origin. This integration, which enables us to applymore sophisticated validation and correlation techniques, should ultimately enhance therobustness of the control system. Moreover, part of the operator's decision making may besupported by proposals issued by this artificial intelligence processing unit.

It is undeniable that faster actions on the ventilation could anticipate more efficiently the up-coming pollution and thus avoid more easily peak concentrations to occur. Indeed, whendangerous pollution levels are measured, it is, from the control point of view, almost too late toreact efficiently. This is because counteracting by increasing, as fast as, possible the air flowrates leads to an even higher consumption of energy. The inevitable time delay between themoment the excessive pollution level is detected and the decision to increase the ventilationlevel is taken, can be shortened by quantifying directly the cause of pollution, i.e. the trafficflow, rather than its consequence, i.e. the dilution of its emitted pollutants in the ambient air. In

237

the latter case, a ventilation level can then be linked in a one-to-one relationship with a set oftraffic parameters.

This idea has been applied in the ventilation control algorithms of the Belliard tunnel. Theexperience so far has shown that this approach works. Of course, this solution does not suggestthat pollutant measurement devices become unnecessary, but their number could be significantlyfewer.

2 OBJECTIVE OF THE PAPER

The objective of the paper is to show, on basis of an extensively commented example of theevery day ventilation management, that a ventilation control system driven by measurements ofcertain traffic parameters has advantages over a system that would normally be based onpollution measurements.

3 BELLIARD TUNNEL

The Belliard tunnel is an urban road tunnel in the centre of Brussels.As may be seen on figure 1, it is a complex tunnel, not only because of its length and its severalaccess and exit roads, but also because of its Y-shape. Considered as a single tunnel, thisparticular Y-shape is due to the assembling of parts of three previously existing tunnels. It linksthe inner ring road of Brussels with the east of the country through the E40 motorway, and withthe south of the country toward the E411 motorway. Moreover, it serves a local area in Brussels,and allows access to the independent underground parking areas of two EC-buildings, viz. theBerlaymont and the Justus Lipsius buildings. With its three-lane entrance, the tunnel is anextension of the important five-lane Belliard street. These three lanes, joined by an additionallane from the Conseil parking lot, form a four-lane road in the first part of the tunnel. Then,after some 460 m, it splits into two separate two-lane roads. The length between the entrance atBelliard street and the most distant exit i.e. the east portal, is nearly 2 km. The southern branchof the tunnel is much shorter and has, for the moment, only a length of 300 m.

The ventilation is of the longitudinal type, but with intermediate air extraction and subsequentfresh air injection. The air flow volumes match at least the fresh air demands in each of thesections of the tunnel. As may be seen on figure 1, there are altogether 13 slots, denoted by Innnfor an injection slot, and Ennn for an extraction slot, nnn being a sequence number. Each slot issupplied by a twin fan system, all of these twenty six fans being equipped with a variable pitchangle system. The total installed ventilation power is around 2.6 MW.

Since the main fans and their corresponding slots in the tunnel roof could not always be locatedwhere a preliminary theoretical study indicated that they should be, ventilation needed to beboosted in some parts of the tunnel by means of jet-fans. These are referred to on figure 1 byGRnn., nn being the number of the group of jet-fans. Since heavy traffic is not allowed in thetunnel, the ventilation has been calculated for an exclusive use by light traffic, of which 10 % isassumed to be equipped with diesel engines. For this reason, possible problems with opacitywere not taken into consideration.

238

The main criteria and standards for calculating the minimum ventilation capacity form thesubject matter of an official ministerial decree, that stipulates that the pollutant concentrations inthe air must not exceed the following values :for CO - 100 ppm as an average value along the tunnel if the exposure time does not exceed halfan hour and;for NO2 - 1 000 µg/m3 as a maximum value if the exposure time is less than 20 min, 400 µg/m3

if the exposure time is greater than 1 hour, and a linear interpolated value for any in betweenexposure time.

239

240

It is also important to notice that the mechanical ventilation capacity in certain sections had to bere-inforced in order to meet the requirements of a minimum air velocity in case of fire hazards.However, this aspect is beyond the scope of this paper.

4 VENTILATION SYSTEM EQUIPMENT

The ventilation system can logically be subdivided into three entities (see figure 2) :1. the supervision unit which is mainly set up by two independent and parallel working

supervision units LS2000, each exchanging data with six programmable data-logging systemsT809 and assisted by an expert system SECAT,

2. the right part which represents the mechanical ventilation equipment and,3. the left part which are all the measurement devices.

4.1 Supervision unit

The responsibility of the government to meet legal obligations requires the set-up of amonitoring system that takes account of the need to obtain representative measurements.Carefully designed programs are also required by engineers who investigate environmentalcontrol technology and energy saving aspects, by operators who run the ventilation process andneed to be well informed about alarms and operational breakdowns, and by rescue people whoare in charge of safety.Since, on the one hand, the control system must operate in real time, the frequency of samplingneeds to be carefully considered. A reasonable monitoring cycle time that comes close to a realtime response in relation to the ventilation and pollution process, has been considered to be1 min. This aspect, of course, determines the minimum processing capacity of the entire system.On the other hand, the system must be reliable in all circumstances, i.e. absolute priority is givento a hardware configuration that in principle never fails, and a software implementation that isfully robust. These requirements justify the set-up of two identical LS2000 working strictly inparallel.

The more sophisticated programs and algorithms are implemented on a single and externalexpert system SECAT. Being permanently in communication with the LS2000, it upgrades thecapabilities of the monitoring system to a control system. Moreover, a data link connects theventilation control system to the central dispatch room from where all the tunnels in Brussels aresupervised.

.

4.2 Mechanical ventilation equipment

Axial fansThe fans, altogether 26, are axial, anti-stall fans with variable pitch angle, each equipped with adamper. They are set up in groups of two fans operating in parallel, each pair being controlledas a single operating unit.

241

The ventilation capacity of each fan has been calculated in such a manner that it corresponds toapproximately 50% of the estimated capacity in case of fire. This solution offers, besidesmaintenance advantages, also the possibility to run alternatively one fan and then the other,mostly at maximum efficiency. The difficulty however, is the transition from one to two fans,and vice versa without disturbing too much the ventilation in the tunnel.

242

243

The single control unit allows one to monitor the different information coming from the fanssuch as heating problems of the bearings, protection against overload of the motor, the start-and-stop procedures, the opening and closing of the dampers, and the pitch angle settings in order tomatch the required air flow quantity.

Jet-fansSince it was not always possible to install air injecting slots where needed, 44 jet-fans wererequired along the tunnel in order to ensure a sufficient ventilation capacity at any one of itssections. They are 11kW fans with a diameter of 0.56 m and capable of an air ejection velocityof 35 m/s. From the functional point of view, they are subdivided into 11 groups. When fullventilation capacity is not needed, this logical subdivision has the advantage that either fanwithin any group can be started up, so allowing to operate them in turn.

4.3 Measurement equipment

Traffic : Flow and speedSince traffic is the source of pollution in a road tunnel, it is obvious, although only recently putin practice, that the measurement of certain traffic parameters has to be considered in the firstplace. The traffic variables, which are the traffic flow volume, i.e. number of vehicles per unitof time, and the traffic speed in km/h, are measured by analysing the successive pictures of thetraffic taken by video cameras. More can be learned in Versavel et al., [1].

However, from the pollution emission point of view, a much more useful information would betraffic concentration, i.e. the number of vehicles per km. Till now, this information has beenobtained by dividing the traffic flow volume by the speed value, but this mathematical operationobviously generates inaccuracies all the more important that the speed drops. Therefore, under10 km/h maximum density is assumed.

Each camera will normally monitor two lanes. The implementation of the cameras is indicatedas TCnn on figure 1.

Pollution : CO concentrationThe CO concentration in the tunnel is measured by making use of the well proven non-dispersive infrared (NDIR) absorption technique, i.e. CO absorbs infrared in a very specificmanner, producing an absorption spectrum. By splitting the infrared radiation into two parallelbeams, one passing through the sample compartment and one to a reference cell, a differentialsignal proportional to the CO level is obtained. This signal is then converted into a standard 4-20 mA output.

Since the principle of the measurement is continuous this measurement device is capable ofoperating on a real time basis. Within the useful operational domain, i.e. between 0 to 200 ppm,the accuracy is 10 ppm or 10% of the reading. This accuracy is achieved by using a cylindricalsample compartment that has a length of no less than 3m.

Of interest is its auto-calibration capability, therefore requiring very little maintenance. Besidesan analogue output signal from 4 to 20 mA, the device signals a faulty functioning and anytransgression of a level setting.

Their location, indicated as CO01 through CO08, may be seen on figure 1.

244

245

Pollution : NO2 concentrationThe sensor put here in operation, is based on the physical principle of chemiluminescence : NOmolecules are oxidised by the chemical reaction between ozone and NO. This reaction gives alight output that is proportional to the quantity NO2 being produced and thus also to the initialNO quantity present.

The real NO2 concentration in the air is calculated by the difference between two measurementsobtained by a double and successive application of the aforementioned principle : during the firsttime, all the NO in the sample is transformed into NO2 and the emitted light is quantified.During the second part of the measurement cycle, the initial NO2 is first reduced into NO andthen the entire NO content is oxidised and again the emitted light is measured. The method isaccurate and reliable only if very low concentrations are to be measured (0 to 1 mg/m3), but itsimplementation as a practical measurement technique results in a very sophisticated andexpensive instrument that must be handled with care. Since they are very expensive, only threeof them have been put in operation. The instruments themselves are not installed in the tunnelbut behind the walls.The actual measurement is done on the basis of samples continuously taken by a small pumpsystem. The full scale measurement range is programmed to be 1 ppm.

Their locations are indicated on figure 1 as NO21, NO22 and NO23.

Air flow volume : In the tunnelThe measurement technique uses two ultrasonic devices, each facing the other on the oppositewall of the tunnel, making a line that is at an angle of approximately 45° to the axis of thetunnel. Each of them emits and receives, in turn, a beam of short ultrasonic pulses. Themeasured difference in travel time between the two beams, is due to the axial velocitycomponent of the air flow, which multiplied by the cross-section is precisely the quantity that weseek to estimate. They allow to measure air velocities, averaged on a straight line, in the rangeof -10 to 10 m/s.

Three of such devices, referenced as QT01, QT02 and QT03 on figure 1 have been installed.

Air flow volume : the fansThe air flow rate through the fans is calculated from the velocity pressure measured at the inletpiece of each fan by means of an electronic differential pressure sensor. Since ventilationcontrol involves only flow variations, this method, properly calibrated, was expected to providesatisfactory results. In fact, due to the manner that most fans have been installed within their ductsystems, the velocity pattern at their inlet is far from uniform and steady. This problem is stillunder study.

Fire detectorsFire detection uses a dual rate-of-rise/maximum-temperature sensor. The sensors are locatedapproximately every 15m along the length of the tunnel. Grouped into so-called zones whichcorrespond roughly to the different theoretically possible fire scenarios, each zone is served byone collective signal line. Therefore, all sensors in the same zone have a common address.

Since this information does not enter into the ventilation control problematic, no further thoughtwill be given in this paper.

246

247

5 CONTROL SOFTWARE

As already mentioned, the objective of the ventilation control system is mainly two-fold :• first, it must provide ventilation systems that are economical in routine operation, i.e.

adapt the ventilation to the real but minimum fresh air needs. Adequate air flow rate isindeed a compromise between two conflicting requirements. On the one hand, the airflow volume must be sufficient to dilute the pollutants emitted by the vehicles. On theother hand, the air flow quantities should be as low as possible to save energy and reducerunning costs. Hence, an efficient control system is the one that, at any time, willguarantee these optimal air flow quantities in every section of the tunnel.

• second, it must control the ventilation as safely and efficiently as possible in emergencycases resulting from breakdowns, accidents or fire. Fires are unfortunately among themost hazardous situations. In this particular case, it is important that the ventilationsystem be capable of controlling the direction of smoke motion in order to provide aclear and safe path for escaping and rescuing people, as well as an easy access for the firebrigade (Heselden, [2]).

From the operational point of view, the software packages have been developed around threeoperating modes :− a "normal mode" which is run by the combination of both the LS2000 and the expert system,− a "degraded mode" which is exclusively kept under the control of one of the LS2000, and,− a "fire mode" which is also exclusively run by either one of the two LS2000.

According to the general meaning of the word "normal", the normal mode is the one that isapplied in all circumstances, i.e. the one that corresponds to the first of the aforementionedobjectives.

The degraded mode is an operational mode with reduced capacities that will be active only incase of deficiency of certain parts of the control system, for example, a partial or complete breakdown of the expert system, an interruption of the data communication links, and so on ...

The fire mode must by definition be very reliable. For this reason this mode runs only on theLS2000. However, since its goals are apart from the one dealt with in this paper, no furtherthought will be given to this aspect.

The underlying difficulties, the control strategy and the subsequent algorithms have beendiscussed in (Jacques, [3]), (Jacques et al, [4]). It has been explained that, as far as complextunnel structures are concerned, the relationship between the required fresh air flow rates at anysection of the tunnel, and the amount of air to be input by each fan is not linear.Indeed, the pollution level in any section that must be taken into account, is not only due to thelocal pollutant emissions of the traffic, but also to the one coming in from the upstream sectionor sections. Another typical feature of complex tunnels is that there is no way to link a particularmeasurement, or set of measurements, to one particular fan system that would have an exclusivecontrol over it.

Moreover, two variable pressure generating sources could interfere with this ventilation system.There are :− jet-fans, some of which could be switched on in certain parts of the tunnel and in other parts

not;

248

− traffic congestion that may occur only in one branch of the tunnel, where simultaneously thepiston effect disappears and the aeraulic resistance of that part rises, while in the other branchthe piston effect is still in effect.

These two elements aggravate the non linear character of the overall process and could inducesignificant ventilation disturbances. In fact, controlling this type of ventilation is not feasiblewithout a reliable numerical model that allows us to evaluate the impact of a disturbance,whether it is accidental or programmed, on the different sections under control. The numericaltool exists (Jacques, [5]), (Ferro, [6]), but are not, in the present state of the art, sufficientlyreliable for control purposes. Yet, the designed control system in the Belliard must not only becapable of boosting the ventilation where the additional pollution due to congestion isincreasing, but must also maintain the economical ventilation in the other parts of the tunnel, orrestore it in those parts where it has been disturbed.

The only way out to solve this control problem and to come out with a workable solution is totry a different approach. The underlying ideas are :− since traffic is the source of pollution, it is not unreasonable to be able to control its

concentration by means of variables which characterise the traffic;− by giving in on the optimality criteria of the control function, it must be possible to subdivide

the operating domain of the controlling traffic variables into a certain number of smallerdomains where these variables could be considered constant.

This procedure allows one to prepare in advance, by means of numerical simulations, a certainnumber of ventilation configurations, each of them being defined in a table which then containsthe set points of all the fans. These tables are addressed by a combination of four, almostindependent variables which characterise the average traffic situation in the tunnel.

The advantages of this approach are obvious :− a table may easily be replaced by a more adequate one as experience with the real ventilation

progresses;− the software is less complicated and thus guarantees a higher reliability and;− any configuration of the ventilation is entirely predictable and can, therefore, be reproduced,

which is particularly useful when testing the software.

Conversely, the weak point of this solution is certainly the lack of guarantee that pollution willnever exceed the threshold limit values. A built in protection mechanism should prevent this.The control algorithm is designed to adapt automatically the sub-domain boundaries of thetraffic variables, if a repeated and significant pollution transgression within a certain time span isnoticed. Pollutant concentrations are thus used as a kind of second level control loop. So, thisapproach can be defined as being centralised, discrete and having a two level control mode :density and speed characteristics of the traffic are used as driving variables, the averagedpollutant concentrations as supervising variables.However, for the sake of robustness of the control algorithm, the second level control loop hasnever been activated, one of the reasons being the weak reliability of the NO2 measurements.This point is still under investigation.

The four control variables are the traffic concentration variables, i.e. the number of vehicles perkm, the average traffic speed, the unbalance between both the exit sections; first where theconcentration is concerned; and second where the speed is concerned. All these variables are, of

249

course, time dependent. Average values are therefore considered. Where the concentration isconcerned, numerical simulations allow us to divide its entire range into five domains : domain1 from 0 to 35 vehicles/km, domain 2 from 36 to 70 vehicles/km, domain 3 from 71 to 105vehicles/km, domain 4 from 106 to 140 vehicles/km, and domain 5 for more then 140vehicles/km. Since speed influences pollution to a lesser extent, only three domains wereconsidered, i.e. 0 to 15 km/h, 16 to 30 km/h and more than 30 km/h. Possible unbalance of thetraffic concentration in both the exit branches, is quantified by the values ranging from -1 up to -0.33 for the first domain, from 0.33 up to 1 for the third domain, the other values coveringdomain 2 where almost equal traffic conditions are assumed. Possible unbalance of the trafficspeed between the two branches are expressed by -1 or 1 if the average speed values aresignificantly different and zero if not.Since these boundary values had to be determined by simulation, they may not represent themost optimal subdivision. In fact, we are convinced that they do not. However, there has beenno need, till now, to adjust these values.

The combination of the four above defined variables yields 135 (= 5 x 3 x 3 x 3) possibilitiesand therefore may address 135 tables. In actual practice, the number of tables is a bit less sincesome combinations of speed and density are not possible. Each of these tables contains thesetting points of all the fans in the system, and thus defines one operating point of the tunnelventilation, which would, at least if the underlying model is simulating the reality properly,satisfy the fresh air requirements in each part of the tunnel. In the case of a systematicdiscrepancy the boundaries of the domains of the traffic variables may, as previously mentioned,be automatically adapted.

In routine, the control procedure goes as follows : every minute the SECAT retrieves all thetraffic measurements coming from the cameras, all of which were previously collected by theLS2000. Since each measurement is provided with validity status information, the systemchecks first whether the number of available and valid data is sufficient. This check isperformed by combining the status information with a weighing coefficient that has beendefined for each camera and that takes into account, among others, the number of lanes seen bythis camera. Summing up all these weighing coefficients allows us to compare the result withthe theoretical maximum value. If the former value is less than half of the latter one, whichmeans that the majority of the cameras fail to work properly, the control system switchesautomatically to a typical pre-recorded traffic diagram according to the moment of the day andthe day of the week.

If enough information is available, the four traffic variables are calculated as a weighedcombination of all the valid measurements, i.e. both the traffic density and speed of the vehiclesat the different location points of the cameras. The overall traffic situation in the tunnel isassessed by evaluating the four traffic variables according to the above mentioned domainspecifications. Still at the level of the SECAT, the most suitable ventilation table is now set upby selecting, on basis of the traffic variables, one out of the 135 prepared tables. Since all theinjecting and extracting fan units are in fact twin fan systems, the expert system determineswhether the required volume flow can be achieved either by one or by two fans running inparallel. If the volume flow demand can be achieved by either one or two fans, the algorithmchooses the solution that optimises either the number of starting or stopping operations by takinginto account some overlap mechanism. The ventilation table built up in this manner istransmitted to the LS2000 whose job it is to put this configuration into operation. However, if

250

within a group one or more fans need to be started up, the LS2000 will always activate them in acyclic fashion. It means that the fans operate one after the other in a circular manner.

One of the real difficulties of this kind of approach are the boundary effects. Indeed, it is almostimpossible to study in advance all possible call sequences of the above mentioned tables.Therefore, it will never be sure that no conflicting situations will occur, e.g. starting up a fan thatwas ordered to stop during the previous control cycle.

However, during the last 3 years that the application ran, we never experienced that kind ofconflicting situation, but this does not mean that they didn't occur.

In lokelihood the SECAT falls out or fails to exchange data with the LS2000, the systemautomatically switches to the degraded version. Tailored on the same ideas, i.e. central anddiscrete, it constitutes in fact a very reduced subset, which means 7 tables instead of 135 tables.The main reason being that the LS2000 is not designed for running elaborated programmes.The simplification is, of course, obtained by reducing the number of independent variables andthe number of considered sub-domains within the remaining variables. Only the traffic densitiesare now used. There are three variables, one corresponding to each branch of the tunnel. Theyare evaluated on the basis of measurements coming from specific cameras in each branch, i.e.cameras TC01 and TC02 in the entrance branch b1 of the tunnel, camera TC04 in the southernbranch b2 and camera TC07 in the horizontal branch b3 (see figure 1).

In order to avoid important fluctuations these values are smoothed out over time. The calculateddensity in branch b1 addresses one out of three levels of ventilation power, i.e. maximum load,and two reduced ventilation loads which are roughly 2/3 and 1/3 of the ventilation capacity. Thecalculated densities in branches b2 and b3 determine, in case of partial ventilation loads, threeconfigurations, i.e. either a relative balanced traffic situation between the branches b2 and b3, or asignificant unbalance with either a predominant traffic flow in branch b2 or vice versa.

Since the degraded mode has been designed only for exceptional situations, it should beemphasised that this operational mode is not supposed to run for a long period, and therefore, theenergy saving aspect has not been of any particular concern.

6 OPERATIONAL EXAMPLE

To illustrate the working and efficiency of the Belliard tunnel ventilation control algorithms, letus describe, with real-time recordings at hand, a situation where the traffic starts to jam.

6.1 Traffic

On the recording of camera TC09R (figure 3), dated 26 July 95, one can see that, between 17:00and 17:15, a traffic incident is about to happen : the vehicle flow leaving the Cortenberg sectionis decreasing slowly and progressively while the number of vehicles entering the tunnel remainsconstant (see figure 4).

251

0

5

10

15

20

25

30

16:00 17:00 18:00 19:00 20:00

TC09Rcar/min 26/07/95

h

Figure 3. traffic flow at exit of Cortenberg (right lane)

0

5

10

15

20

25

30

35

40

16:00 17:00 18:00 19:00 20:00

TC03Rcar/min 26/07/95

h

Figure 4. traffic flow at entrance of Cortenberg (right lane)

After 17:14, one can read from the same recording of camera TC03R (figure 6) that the trafficstarts also to congest at the entrance of the Cortenberg section. Within 3 min, the average speedhas dropped from 77 km/h to 40 km/h, and then further still to 29 km/h. At 17:54, at the samelocalisation, the traffic jam has suddenly disappeared, and within 3 min the traffic speed hasincreased from 27 km/h, to 33, and finally up to 62 km/h. The traffic flow on this lane has nowdropped from 25 vehicles/min to 15 vehicles/min.

At 17:50, and until 18:01, at the exit of the Cortenberg section, on the left lane, the cameraTC09L (figure 5) shows that the traffic flow has suddenly increased from 11 up to 30vehicles/min, while it has been interrupted for 2 minutes on the right lane (see figure 5). Thisshows the end of the incident.

252

0

5

10

15

20

25

30

35

16:00 17:00 18:00 19:00 20:00

TC09Lcar/min 26/07/95

h

Figure 5. traffic flow at exit of Cortenberg (left lane)

Resumption to the normal traffic speed in the tunnel doesn’t take long coming : between 17:53and 17:56 it goes up from 18 to 62 km/h at the splitting of the sections Schuman and Cortenberg(see figure 6).

253

0

20

40

60

80

100

120

16:00 17:00 18:00 19:00 20:00

TC03Rkm/h 26/07/95

h

Figure 6. Traffic speed at splitting of Cortenberg and Schuman

The development of a traffic congestion can be explained using the theory of attractors : thetraffic speed changes almost instantaneously from one value to another. A minor incident maytrigger this phenomenon, in the same way that a slight opposite reaction could make it totallydisappear.

6.2 Ventilation

At 17:16, the new traffic situation seems to be identified by the expert system. In order toanticipate the potential risk of an higher pollution level, it is necessary to adapt the ventilation.Therefore, the new values of traffic concentration and speed are used to determine the address ofthe corresponding ventilation table, the content of which is then sent to the control LS2000. Thelatter takes charge of dispatching the new setting points of all the fans involved by activatingtheir local control units. Four minutes later, i.e. at 17:20, the new ventilation level is achieved.

Between 17:50 and 17:55 the traffic incident is closed. There is no more need for furthermechanical ventilation assistance, and thus the fans are switched off. This intervention is clearlyillustrated in figure 7, which shows the air flow volume of one of the fans involved, i.e. theinjecting fan I015 (see figure 3).

0

10

20

30

40

50

60

70

16:00 17:00 18:00 19:00 20:00

QI15m3/s 26/07/95

h

254

Figure 7. Air flow volume of fan I015

255

6.3 Air flow volumes

0

1

2

3

4

5

6

7

8

9

16:00 17:00 18:00 19:00 20:00

QT03m/s 26/07/95

h

Figure 8. Air velocity at the Cortenberg exit portal

The velocity of the air flow recorded by the ultrasonic device QT03 (see figures 1 & 8) locatednear the exit portal of the Cortenberg tunnel, is decreasing gradually between 17:02 and 17:11.Clearly it is due to the piston effect of the traffic that slows down from 6 to 3 m/s. After theventilation has been started up, at 17:20, the velocity went up again to a value of 6 m/s.

After 17:50, the velocity increased up to 9 m/s, then dropped to its normal value at 18:05. Thisis explained in the first moments by faster traffic and later on, by the switching off of themechanical ventilation.

6.4 Pollution

Up to a certain point the growth of the pollution is « normal ». This is due to the increasingtraffic leaving the city at the end of the afternoon. This is also true for the CO concentration,recorded by the CO07 meter (see figure 9), as well as for the NO2 concentration, measured bythe NO23 meter (see figure 10), both sensors being located near the exit of the Cortenbergsection.

0

10

20

30

40

50

60

70

16:00 17:00 18:00 19:00 20:00

CO07ppm 26/07/95

h

256

Figure 9. CO concentration near the Cortenberg exit portal

0

100

200

300

400

500

600

16:00 17:00 18:00 19:00 20:00

NO23ppb 26/07/95

h

Figure 10. NO2 concentration near the Cortenberg exit portal

Then, the local pollution concentration at the same location, while remaining small, indicates thebeginning of a traffic problem. The pollution increase at the entrance of the Cortenberg tunnelonly appears at 17:19. This increase is mainly due to the reduced air flow volume by a smallerpiston effect, rather then to the increasing traffic flow.

When the traffic jam starts to develop, the expert system expects an excessive forthcomingpollution and decides to start up the mechanical ventilation. As a consequence, the air velocityin this part of the tunnel is boosted up, so compensating the lack of piston effect. The COconcentration decreases towards the exit while it remains unchanged in the entrance part of theCortenberg section. In the case of NO2, the concentration fades away significantly after theventilation has been switched on.

7 CONCLUSIONS

A few important conclusions can be drawn from our 3 year experience with the automaticventilation control system operating in the Belliard tunnel :

The system seems to work well, although we must admit that there are only very few cases oftraffic congestion, on average, once a week. It turns out that most of the time the controlalgorithm runs only with three tables. This would mean that the combinations of the trafficvariables, speed and density, are closely related. This fact has indeed been observed and is nowbeing investigated more thoroughly. It would also be worthwhile to identify these mostfrequently used tables in order to push further their optimisation.

The action on the ventilation would not have been as fast, if the measurements of the COconcentration, and respectively the NO2 concentration, were used instead of the trafficinformation. The example described in this paper shows it clearly. As explained, this fact is dueto the design of the ventilation control system that reacts on significant changes of the trafficflow which is the generator of the pollutants, rather than on variations of its consequence, i.e. thepollution in the tunnel, which is further more dependent on the ventilation itself.

257

Although it is difficult to prove, we learned from our 3 year experience that the ventilationcontrol is less sensitive to measurement deficiencies in case of a traffic based control systemthan it is in case of a pollution driven control system. This point is important if one knows,particularly where complex tunnel systems are concerned, that equipment failures and wrongmeasurement information are rather the rule in every day life of the tunnel, than they are theexception.

8 REFERENCES

1. Versavel, J., Lemaire, F., Van Der Stede, D., Maes, W. & Mortelmans, J., ‘CCATS’ -Camera and computer aided traffic sensor, CCCT '89 Conference, IFAC, Paris, France,1989, pp.63-68.

2. Heselden., A.J.M., Studies of fire and smoke behaviour relevant to tunnels. Proc. 2ndI.S.A.V.V.T., BHRA Fluid Engineering, Cambridge, England, 1976, paper J1.

3. Jacques, E., Study of the automatic control of the longitudinal ventilation of a complex roadtunnel. 8th International Symposium on Aerodynamics and Ventilation of Vehicle Tunnels,Mechanical Engineering Publications Ltd., London, England, 1994, pp.943-957.

4. Jacques, E., Possoz, L., Road tunnels and the environment. Proceedings of a workshop onMeasurement techniques in energy systems and processes, Tempus Structural JointEuropean Project S JEP-07397-94, Niedzica-Gliwice, 1996, pp.83-108.

5. Jacques, E., Numerical Simulation of Complex Road Tunnels. Aerodynamics andVentilation of Vehicle Tunnels, Elsevier Science Publishers Ltd., England, 1991, pp.467-486.

6. Ferro, V., Borchiellini, R. & Giaretto, V., Description and application of a tunnel simulationmodel. Aerodynamics and Ventilation of Vehicle Tunnels, Elsevier Science Publishers Ltd.,England, 1991, pp.487-512.