S.Sjöström*,E.Klintenäs,P.Johansson,J.Nyqvist

Department IAPI– Mining Automation,ABB AB,Umeå 1400,901 24,Sweden

Keywords:Mine ventilation Ventilation on demand Optimized model-based control Minimized energy consumption Advanced process control

ABSTRACT In early 2018,the Boliden Garpenberg operation implemented an optimized control strategy as an addition to the existing ventilation on demand system.The purpose of the strategy is to further minimize energy use for main and booster fans,whilst also fulfilling airflow setpoints without violating constraints such as min/max differential pressure over fans and interaction of air between areas in mines.Using air flow measurements and a dynamical model of the ventilation system,a mine-wide coordination control of fans can be carried out.The numerical model is data driven and derived from historical operational data or step changes experiments.This makes both initial deployment and lifetime model maintenance,as the mine evolves,a comparably easy operation.The control has been proven to operate in a stable manner over long periods without having to re-calibrate the model.Results prove a 40% decrease in energy use for the fans involved and a greater controllability of air flow.Moreover,a 15% decrease of the total air flow into the mine will give additional proportional heating savings during winter periods.All in all,the multivariable controller shows a correlation between production in the mine and the ventilation system performance superior to all of its predecessors.

1.Introduction

The Boliden Garpenberg underground mine operation have been taking a phased approach and deployed ventilation control in stages [1].They have run ventilation on demand successfully since 2008.The ground principle of current ventilation on demand systems is each cell of the mine having a set of secondary fans in different working areas.Activities,in demand of air,are detected in the working areas using location tracking services.Corresponding fans are set to operate at a setpoint determined by the type of activity detected.Aggregation of the total secondary air flow needed for each cell is calculated and used as air flow setpoints for the booster fans on respective level.Main fans are controlled by differential pressure controls.

Many mines completely lack a control system,and even the state-of-the-art technology for control,ventilation on demand,VoD,has its drawbacks:there is no feedback control and the fan/air regulator relation models are weak and complex [2].

As a way to achieve an energy-optimized and reliable solution that automatically feeds the mine with the required air,ABB has developed and now offer a new,unique method for mine-wide coordinated control of the fans and air regulator.The solution is based on empirical models and relies on feedback from air sensors,for instance gas,flow or temperature sensors.Multivariable models describe how changes in the speed of fans affect both the airflow and the pressure over fans.The parameters in the models are obtained empirically from operational data which makes the model easily adaptable to new conditions.

A fingerprint of the ventilation system indicated an improved coordination between the ventilation objects that would enable further energy savings.The new Advanced Process Control method was suggested to be implemented as a complement to the current ventilation-on-demand-functionality.This paper describes the implementation of this control method in the Lappberget area of the Garpenberg mine (Fig.1).A process view of the Lappberget mine ventilation control system is shown in Fig.1.

2.Methods

2.1.Control definition

The ventilation system in the Lappberget area of the Garpenberg mine utilizes multiple fans to distribute the air flow.Doors are used to access the three different mine levels.All fans and doors are part of the same ventilation system and any operational change of one singular fan or door will have operational impact on all the other fans.The control problem will be difficult to accurately solve using SISO (Single In,Single Out) controllers [3].

Fig.1.Ventilation system layout in the Lappberget area.

The target of the control is to operate the fans in a coordinated manner,fulfilling the required minimum flow of air at all levels and creating safe operation points for the fans,using as little energy as possible.

2.2.Mine wide coordination model

The model used to describe the mine ventilations system is inspired by dynamic multivariable models [4].Such models are widely used in many applications,eg.,in the refining industry.However,substantial efforts are normally required to obtain the models [5].For the situation at Lappberget,this is undesired since the layout of the mine continuously changes through expansions and accidental new paths of air flow created as a result of blasting.In this case,a simple static multivariable model,which is able to capture the essential interaction and impact from changes in the actuators,is used.The model is described in the incremental form as follows:

where Q is the vector of measured airflows;p the vector of pressure over fans;β the vector of fan speeds;E the vector of fan powers;and Δ changes between two samples.

The coefficients in the matrices Hq,Hp,and Heare obtained from simple experiments,or from normal operational data that enables the automation of system identification[6].The initial first tests of the model are presented in [7].

2.2.1.Measurements

To identify the model of the mine ventilation system and to be able to use it for control,a number of variables must be measured.The variables are gas concentration and/or air flows in various locations to be controlled,fan power of the fans to be controlled,fan speed of the fans to be controlled,pressure over fans.The controller adjusts the actual fan speed based on current demand.

2.2.2.Control

The objective of the control system is to maintain the desired airflow in various airways,while minimizing the power required to run the fans.Airways that feed production areas with air must have an airflow that exceeds the required flow.Other airways may be requested to have airflows which should be kept as small as possible.This can be formulated as an optimization problem where new fan speeds are determined to minimize the actual fan power.

During minimization,constraints on airflow and on differential pressures are taken into consideration:

In addition,there are limitations on the fan speeds:

This optimization problem is solved on a cyclic basis where new fan speeds are calculated.The initial values for each optimization are provided by filtered values of the measured signals.

2.3.Defining the control scope -field fingerprint

A fingerprint was done to investigate possibilities of additional energy savings.An analysis of the ventilation system was done to identify such possibilities and to set the control scope properly.Site visits and interviews with mine ventilation and automation specialists were performed,and the system limitations and potential for improvement using an advanced control approach were identified.The analysis resulted in an improved and superior view of the ventilation system in which the secondary room fans are abstracted.The focus was put on the main flows as shown in the left part of Fig.2.The fingerprint analysis identified three main areas of interest:the target values to aim for,the parts of the mine involved in achieving them and the conditions constraining the control options.The analysis also showed a need of additional differential pressure sensors to be installed.

Fig.2.A Schematic sketch of the main ventilation system in Garpenberg mine.The secondary fans providing air to the production areas (black boxes for illustration)are not visible in the sketch.

The parts with most influence on the main air flows were found to be the main fans and the pair of intake and exhaust booster fans on each level.The secondary fans on each level act as agitators only and have no significant impact on the overall flow of air.

The fingerprint investigation showed that the two most important target values to control in order to achieve maximal energy savings were the amount of the air flow in and out of each level,as well as minimizing the fan power needed to achieve these flows.The target flows are calculated by the ventilation on demand functionality (Fig.3,framed in green).

In addition to the target values,there are also several constraints:leakages between the levels need to be controlled and kept under a certain limit and the differential pressures over the fans need to be kept on the positive side to make sure that the fans are operating within their defined working area functionality(Fig.3,framed in yellow).

The measured signals,shown in Fig.3 above,used were fan speeds of the fans to be controlled (framed red),fan powers of the fans to be controlled (framed red),air flows in each cell to be controlled (framed green),air flows between the cells and as constraints(framed yellow),air flow at the lowest level out of the system as constraint (framed yellow),and differential pressure over each controlled fan as constraint (framed yellow).

2.4.Model determination -system identification

To obtain the coefficients Hq,Hp,and Hein the model formulation,a system identification was carried out.The model of the ventilation system used for control of the fans was identified through a series of step change experiments,performed live on the site system in coordination with the mine operator.To best capture the impact on each control input (fan speed) on the complete system,all automatic controls of the involved fans were first turned off.Then each fan speed,one at a time,was given a step input and the system was allowed to reach an equilibrium.When stability was observed for all monitored signals:airflow,differential pressure,and fan power,the result was recorded,and the fan speed was returned to its previous level,before the process was repeated for remaining fans.The result of a typical step change sequence is shown in Fig.4.The graph at the top shows the step changes of the fan speed,the graph in the middle shows the response of air flows and the graph at the bottom shows the response of the differential pressure.

This process was performed on all fans in the system,and all the experimental data was combined into a single numerical model of the system.Since the implementation makes use of the ABB Ability 800 × A APC functionality,the APC Model Builder was engaged in doing so.

2.5.Control implementation

2.5.1.Instrumentation extension

The results of the analysis showed a demand for actual measurements not installed in the mine at the time of the analysis.Apart from some of the fan differential pressure measurements,the missing instrumentation was added and installed in the mine as a part of the project.

2.5.2.Model

The relationships in the ventilation system are simplified in the model as linear correlations.The results are shown in a linear approximation of the behavior of the system; the effect of each control input to the system on each of the modeled variables.The model is formulated so it can be implemented in an MPC(Model Predictive Control) controller framework [4].The result is an MPC controller with a simplified model of straight gain coefficients,i.e.no derivative states are modelled.

This simplified approach sacrifices some controller speed compared to an extensive dynamic MPC model,but with the benefit of simplicity and has been proven to be robust for these kinds of cases[7].

2.5.3.Model execution

In the target system,the APC execution engine resides in a dedicated application server.The APC execution engine is represented in the regular control environment of the system,in which it is interfaced to the relevant parts of the already existing ventilation on demand application.The execution of the model controller is iterative,with a step time of around 10 s.In each iteration,the real time values of process signals,target values and constraints are read by the controller.It then uses the internal numerical model to calculate the set of possible control moves (changes in fan speeds) to realize the constraints and target setpoints (min/max/setpoint airflows,min/max differential fan pressures),and use the set that would result in a minimal total power consumption for the complete set of fans.The newly calculated fan set points are forwarded to the ventilation on demand application as set points for the fans.After the iteration is complete,new sensor values are collected and a new iteration is started.

All control moves are constrained to prevent too large control movements which might introduce oscillations in the system given the simplified linear nature of the model.The model is tuned to achieve best possible performance.An example of a result is shown in Fig.5.The graph shows how well the model follows a reference air flow (green) in cell 1 compared to measured air flow value in cell 1 (purple) and the corresponding changes of surface fans(blue),inlet fans to cell 1(light blue)and outlet fans cell 1(orange)over a 24-hour period.

2.5.4.Integration with subordinate ventilation control

Fig.3.Abstracted ventilation model -target values in red and green and constraints in yellow.

Fig.4.Step response and results.From top:Fan speeds; Airflows; Differential pressures.

Fig.5.Reference air flow,measured air flow and speed of nearby fans.

As mentioned above,the signal interface to the model execution services is made available as a function in the regular control application environment of ABB Ability System 800 × A (Fig.6).The actuator to control must have the resulting setpoints from the superior control available and the superior control must be presented with real time values such as fan speeds and flow values(Fig.5).Both in reliable ways.

In the Garpenberg mine,the object control for the fans and instrumentation is within the ABB environment,and the integration with the APC control is implemented using standard application engineering tools by connecting them together.From user interfaces,the integration allows the user to decide whether to run the fans according to the APC control or using the normal ventilation on demand control.Information about which fans are not in ‘‘APC control mode” (either by operator action or i.e.equipment breakdown) is returned to the controller.The corresponding output signals can be put in tracking mode or,if desired,off.The entire APC controller can be switched off and all fans reverted to the previous control mode.This approach makes commissioning very smooth (Fig.7).This also ensures that any safety function in place in the subordinate control,how to control ventilation during a fire for example,will have precedence over any of the superior optimization controls and ensure a safe operation at all times.

3.Results

The air distribution in the mine has been proven to satisfy the demands to a greater degree than ever before.The energy reduction is a direct result of the fans operating in harmony for optimum result.The fact that it is a single system simplifies observations of connections between the control adjusting the flow of air in one area and the common result of air flowing a bit over setpoints in nearby areas.Controllability does not allow for exact proportional distribution with the existing mine design,but the important part is that the air flow setpoint has not been violated on the low end,leaving areas short of air.The results prove a 40% decrease in energy use of the fans involved.Moreover,a decrease of the total air flow into the mine of about 15% can be observed.This yields additional,proportional,heating savings during winter periods.

Beside the direct fan energy reduction that it provides,it also enables a greater controllability of the air flows.This means that the demands,as calculated by the ventilation on demand functionality,can be calculated closer to the limits.If the setpoints are at risk of being violated,some safety factor in the demand parameters are commonly found.With this new system,these can now be reduced or eliminated,see Fig.8 for illustration.It also means that it is easier to avoid conflicting constraints coming into play.The handling of the leakage flows is an example of an entity which was difficult to keep close to its constraints without oversupplying air compared to the actual demands,prior to the new control.Now,it is simply a matter of adding the constraints and setting the limit.

The control has also been proven to operate in a stable manner over long periods without any need of model re-calibration.This while controlling the air flows under all possible combinations of working conditions and despite the unavoidable actual change of physical layout of the ventilation areas.Fig.9,below,shows the total power consumption during a period of 10 months.The graph then shows how a controller was turned on.One can notice that the power slightly increased at the last part of the period.This is due to one of the flow sensors not working properly,resulting in one the fans going on full speed all the time.

Fig.6.APC implementation in ABB Ability 800 × A.

Fig.7.Example of invocation and setup of APC model control directly in the control system (Left:operator view; Right:configuration view for signals used by the APC).

Fig.8.APC (MPC) model control result patterns.

Fig 9.Total fan power of the controlled fans over a time period of 10 months.The vertical line shows when the controller was turned on.

Fig.10.Ventilation system performance KPIs monitoring the tested control system method.

During the same time period,a ventilation performance service monitored the performance of the ventilation system through a set of new KPIs.The service indicated that the increase in fan power was a result of lower performance of the ventilation system.This initiated a fault analysis of the ventilation system which resulted in a repair of the broken air flow sensor.Fig.10 shows a diagram of the ventilation performance KPI over same period of time as in Fig.9.This is an ongoing research and will be described in further detail in another report but is out of scope for this article.

All in all,the optimized control shows a stronger correlation between actual production in the mine and the ventilation system performance than ever before.

4.Discussion

The results of the optimization efforts are according to,and beyond,expectations.The initial target of energy savings was met,and the greater controllability of the air flows opens new possibilities to drive the control even closer to design constraints.The fingerprint and analysis of the ventilation system was imperative to finding the proper scope of control and instrumentation prior to starting implementation.Implementation went hassle free as the scope and expectations were all already in place from the start.

The fact that the model is numerical and not analytical means that it is data driven and is derived using historical operational data collected or experiments involving step changes.This new approach means that both the initial deployment of the control,as well as lifetime model maintenance,as the mine evolves,is a comparably easy operation.The control has also been proven to operate in a stable manner over long periods without any need of model re-calibration.However,sooner or later,the changes of the mines will take its toll on the optimization gains and it is therefore important to monitor the outcome continuously to find the proper timing for model re-calibration.

5.Conclusions

Applying these dynamic empirical models to mine ventilation systems in order to optimize the overall air distribution brings multiple benefits.Not only does the mine owner or operator benefit from reduced running costs,but the working environment benefits from receiving optimized air flows that best fit the activity in that location.Despite the evolving nature of the mine,the existing model keeps pace,ensuring that air and fan optimization is continuous.

(1) Integrated tools and ease of use makes a difference

With advanced control seamlessly integrated into control system environment,the user can easily change the interface signal and tune constraints without additional knowledge or tools.As a result,the historically quite normal fate of model control; slow deterioration in control performance and becoming obsolete over time,will not occur in this instance.The ease of configuration furthermore sets the stage for continuous improvements,allowing analysis of the current situation to turn into actions in updated controls without delays or major cost.

Fig.11.The working methods of evaluating,implementing and sustaining.

(2) Defined work processes set common expectation space and sustainable performance at focus

The method of working out the possibilities and the value proposal of the basic design during the fingerprint analysis is recommended.It is also a part of a work method commonly used for advanced control implementations in various applications.The evaluate-implement-sustain cycle proves to work also in this implementation (Fig.11).

It is still too early to come to any firm conclusion regarding how to best determine when sustaining actions are to be taken on the ventilation optimization implementation described in this paper.It is however obvious that automatic monitoring of the performance of the control is desirable to sustain the savings as the mine evolves craves an alertness that is suitable for automatic monitoring and the capability of indicating the reason of the results going down.Whether it is the result of changes of the mine layout on a scale craving model re-calibration,malfunctioning equipment and sensors or other reasons of non-optimal results,performance monitors should indicate accordingly.Moreover,the continuous monitoring serves a good input for ventilation management programs in aiding in the process of increased operational effectiveness,increased safety,improved engineering efficiency,reduced appliance(fans,doors,regulators,ducting,etc.)maintenance costs and reduced energy costs [8].Such monitoring functions for the ventilation performance is currently under further development.

(3)The importance of different implementation layers in ventilation controls

The control implementation in this instance is layered in the sense that even without the optimization control described in this paper in operation,safe and stable ventilation function is sustained by the lower control layers.This indicates that cloud-based instantiation of optimization functions is a real possibility without compromising production stability.This in turn opens new possibilities of ease of deployment and control sustain actions using emerging technologies such as machine learning.It also works in favor of closer co-operation between different players in the ecosystem.Leveraging existing and new technologies to enhance the way we impellent and sustain benefits throughout lifecycle of plants will move such optimization functionalities closer to autonomous operation.

The result of the optimized control implementation described in this paper,yielding the savings results in stable and safe manners,all in all shows a stronger correlation between actual production in the mine and the ventilation system performance than ever before.The next step in Garpenberg is to extend the model-based advanced optimization control also to the remaining parts of the mine.