Thickeners multivariable predictive control: Division Codelco Norte Guillermo Cortés Codelco, Chile
Jaime Cerda Kairos Mining, Chile
ABSTRACT The mining industry gives central importance to rational and efficient use of water in their operations by adopting measures to optimise their consumption through better management practices. For this reason, Honeywell’s Advanced Control Applications, together with Codelco Norte, have implemented advanced control strategy in High Rate thickeners. The advanced control technique used for implementation is the so called Model Predictive Control (MPC), which aims to drive the continuous process variables under normal operating conditions. Profit® Controller or Robust Multivariable Predictive Control Technology (RMPCT) is Honeywell’s multivariable control and optimisation application for complex and highly interactive industrial processes. The overall process model is composed of a matrix of dynamic sub-process models, each of which describes the effect of one of the independent variables (Manipulated Variables and Disturbance Variables) on one of the Controlled Variables. A sub-process model describes how the effect of an independent variable on a Controlled Variable evolves over time. The strategy objective is “Stabilise and govern the process of thickening, maintaining operational range and ensure proper operation of the thickener.” To reach that objective, the matrix multivariable control solution manipulates the flocculant dose and opening of the valve-discharge as manipulated variables to keep in range: level interface, mechanical torque, underflow solid average percent. The RMPCT will be an effective tool for operations support, helping operators to attain their own production commitments and to have controlled a process that can take up to nearly two hours to see a response to a set-point change. The results obtained with RMPCT respect to manual operation are: (a) Underflow solid average percent has been reduced by 22.3% in variability but with a similar mean value, (b) Mechanical torque has been reduced by 35% in variability and 29.2% in mean value, (c) Level interface has been reduced by 37.57% in variability but with a similar mean value, and (d) the use of the strategy saves 2.8% of fresh water.
INTRODUCTION
The Concentrator of División Codelco Norte, Chile has nine tailings thickeners, which are used to thicken the tailings tailings and reclaim reclaim water from them. them. Several control strategies, strategies, including including expert control, have been implemented in the thickener, but the results were not satisfactory, therefore the thickening process continued to run in manual operation. – 1 –
The purpose of this paper is to present what has been the experience and some results obtained when implementing a new control approach. This paper is structured as follow. Section describes the thickening process and the available equipment. The control objective is shown in Section 2. The control approach is described in Section 3 while Section 4 presents different aspects of the implementation. Results and discussions are presented in Section 5. The paper finalises with the conclusions as presented in Section 6.
PROBLEM STATEMENT
This section presents a brief description of the thickening process as well as the available measurements of the variables and control loops to be used for implementing advanced control strategy. It also include a brief description of the operational parameters and the purpose of the control system. Thickening process
The concentrator at Division Codelco Norte, has nine Tailings thickeners, which are used to thicken the tailings and reclaim water from them. The T7 and T8 thickeners have a diameter of 325 are a high-tech edited capacity by Door Oliver with a processing capacity of 50904 TMSD with a drive unit to withstand a torque of 4800 Klbs-ft. The High Rate Thickener is a cylindrical-conical tank; its function is to reclaimed water from the tailings of the flotation process. The thickener has a canal of incoming feed, a central ring, a system of central drive rake, a peripheral canal to retrieve clear water and three solid discharges in the centre bottom of the thickener, as shown in Figure 1.
Figure 1 High Rate Thickener: Side view and top v iew
Feeding sinks before reaching the centre ring of feed (feedwell) and injected with a nozzle, the clear water of the surface of the thickener is absorbed by Venturi effect which dilutes the pulp fed. In feedwell, the feed is separated into two tubes where flocculant is added in four points to each tube, to stimulate flocculant / pulp mixture and thus enhancing the flocculation of the solids fed.
– 2 –
The solids settle to form a bed, with solids concentration increasing towards the underflow at the base. Clarified water is collected at the peripheral overflow, while a centrally driven rake promotes consolidation and assists sediment transport to the underflow. The rake system consists of two primaries and two secondary rakes, with a hydraulic mechanism of central traction. The rakes assist in dewatering sediment within the thickener bed and moving slowly so as not to cause turbulence and to transport the sedimented material to underflow in the centre of the conical bottom. Control loops and measurements
Figure 2 depicts a diagram of the High Rate Thickener 7 with the available measurements and control loops.
Figure 2 Control diagram of high rate thickener T7
Flocculant adding
The flocculant addition is proportional to the tonnage treated in the thickener (calculated using the measurement of flow in the canal and assuming a percentage of solids in the tailings measured by sampling food or laboratory) and water is added to the flocculant pipe in proportion to the flow of flocculant. Measuring Water-pulp interface
In all tailings thickeners measurement of pulp-water interface is done by a sonic sensor immersed type supported on the bridge of the thickener.
– 3 –
Underflow Solid average percent
Each discharge line has density measurement using nuclear density meter and a control valve type "muscle" with hydraulic actuator with optional remote operation of DCS, which allows adjusting the discharge flow. Hydraulic units for muscle valves are in the "Hydraulic Room", which also contain instrumentation panels and density transmitters. Remote operation of the valves is satisfactory. In the High-Rate Thickener T7, the opening and closing of a valve cutting is done by a hydraulic actuator operated locally and fed by a hydraulic unit with quick coupling system allowing you to use the same hydraulic unit to operate the three valves knife. Reclaimed water
Reclaimed water from the thickeners, obtained by overflow, drain to a peripheral channel, then leads to a rectangular canal by passing it through a sinkhole. Then it is transported to a pumping station to return it to the grinding plant. There is no measurement of flow in reclaimed water collecting gutters which implies mass balances in imprecisely. Operational parameters
The operational parameters of the High Rate Thickener T7 are: Mechanical torque: Less than 30% of the Nominal torque • Flocculant dose: 2 grammes / ton to 4 g / t Normal, 6 grammes / Ton Max • Discharge flow: According to the% solids to be taken, it must maintain balance, so that • what goes in equals what comes out Level interface: 1.6 to 2.0 metres • Underflow Solid average percent: 58-60% according to design • Pulp flow: 2985-5220 m3/hr. • Food tonnage TMSPH:
[email protected]% solids by weight, Max. • Process control objective
The principal issue in the thickeners is related to the type of control loop used for the operation. This is illustrated for the typical actions executed by the operator: the operator entered a setpoint for the flocculant dose, % average discharge opening and he waited one or two hours to see the responses in the Underflow Solid average percent, the leel interface and the mechanical torque. These responses could show errors and a not desired behaviour. From this illustration it is clear that this multivariable process operated in open loop; and it was not possible to rectify any error that could be present. It could not also compensate for the effects of disturbance in the system Therefore, to stabilise the process and guarantee a proper operation of the thickener it was necessary to modify the control approach where the process control objectives are: Reduce the standard deviation of % underflow solids and mechanical torque • Keep the equilibrium condition of the thickener • Reject the effect of disturbances originated in the upstream milling and flotation process • Prevent high torque in the rake • Keep the Differential Inventory Accumulated Charge into a range •
– 4 –
The feasible control strategy should satisfy the objectives stated above, if that objective cannot be reached directly it should, at least, point even indirectly and implicitly to it. The use of the controller will reduce the standard deviation of critical control variables, which will improve the performance of thickener, maximise the Underflow Solid average percent, keeping the torque controlled without to damage the thickener. Indirectly the controller will improve reclaimed water. Advanced control strategy selection
Thickener application has been implemented using technology called Profit Controller™; this is a Multivariable Predictive Control algorithm based on models. This tool incorporates a dynamic model of the process, comprising a transfer function matrix. This controller, by knowing the future behaviour of the controlled variables (inputs to the controller), can determine the best set of values of the manipulated variables (controller outputs), which are sent as set points to the PID controllers used in the process. Figure 3 depicts the MPC scheme, where k is the current time, H is the time horizon, CV and MV stand for the input and output variables, respectively. At current time k, based on the model of the plant, a sequence of the optimal control inputs MV(k), MV(k + 1), ..., MV(k + H) is computed by solving an optimisation problem over the prediction horizon H meanwhile satisfying input and output constraints. But only the current one is applied to the plant. The optimisation is repeated in each time point, i.e. k + 1, k + 2, .. until CV(k) reaches the High Limi.(For this case, where High Limit is the optimal).
Figure 3 MPC scheme
For this reason we have selected a Multivariable Predictive Control (MPC), which contains a structure that allows effective management of interactions of the process thus enabling its action to reduce the variability of the operation, given the greater capacity of governance the process. These are control algorithms that use an explicit process model to predict the future plant response. According to this prediction in the chosen period, also known as the prediction horizon, the MPC algorithm optimises the manipulated variable to obtain an optimal future plant response. The input of chosen length, also known as control horizon, is sent into the plant and then the entire sequence is repeated again in the next time sample.
– 5 –
IMPLEMENTATION PHASES
The methodology for the implementation of a controller has three important phases that can be recognised during implementation; viz. Test Protocol, Model Identification and Controller Implementation. Test protocol
Process model identification is based on input/output performance data, generated by test signals. Test signals are required to excite both steady-state (low frequency) and dynamic behaviour (high frequency). Test protocol is then the most important task during the implementation of a multivariable controller. Most of the implementation time is consumed during the data generation task. In this context selecting the best signal generator is an important decision, in order to provide rich data for the identification phase. In this sense there is a need to select the signals providing the largest “power spectrum” (rich in frequency) in order to provide a good data set for identification. Because of its power spectrum features, PRBS (Pseudo Random Binary Sequence) is recommended to be used as signal generator rather than step or ramp signals as shown in Figure 4. To prepare a formal plant test, a pre-test is usually necessary for the following reasons: Improve instruments calibration and filtered signal • Evaluate PID performance for each MV(Manipulated Variable) • To obtain steady state time for each CV(Controlled Variable) • To obtain models and relationship for each DV(Disturbance Variable) • To obtain settling time • To obtain data for initial identification •
Figure 4 Profit Stepper system
Model identification
A fundamental problem for any controller is the choice of the model that should be used to represent the system. In general, the model is one of the following: Linear time-invariant (lumped parameter), Linear time varying (lumped parameter), Linear with distributed parameters or Nonlinear. – 6 –
System identification remains both an art and a science. The science is concerned with parameter estimation; the art is usually concerned with determining structure/order, the excitation requirements, and accuracy. System identification involves two steps. First, a sequence for exciting the process to be modelled is specified. A family of candidate models is then proposed. After this a representative member of this family is selected. The second step is a parameter-estimation problem. Parameter estimation is basically the determination of the best set of candidate model coefficients such that the model represents the causal input/output relationships. The identification process is executed in PDS that provide a Matrix of models; each sub model represents the dynamic relationship between MV/DV and all CV’s affected. Because all sub models are linear and time-invariant, superposition principle applies. The obtained models are used by the controller in order to predict the behaviour of CV’s. Profit Controller calculates the actions needed to drive back each CV into its control range. The generated MV actions are based on the following optimisation criterion:
min
2
u , y
W ( Au − y real ) ,
(1)
Subject to:
ROC min
≤ Δu ≤ ROC max
MV min
u ≤ MV max
CV min
≤
≤ y ≤
CV max
(ROC Rate of Change of MV within ROC limits) (MV’s lie within high and low bounds) CV’s must lay within high and low bounds)
∧
where, y (the Predicted CV) can be represented as: ∧
y
=
Au
,
where A is the process model and u is a set of MV moves. The solution “u” is the control action and the solution “y” is the optimal CV response trajectory. Control strategy implementation
The advanced control strategy of a thickener process has to achieve the following performance objectives: (1) safety/operation constraints; (2) reducing the standard deviations of controlled variables; (3) de-bottlenecking the process; and (4) maximising product value (or profitability). The objectives are prioritised in this same order, but the order can vary from operation to operation within an application. We use step response modelling for the MPC controller, proven to be effective in Profit Controller applications. Step response modelling makes prediction of process outputs available explicitly. Future prediction is used to compute the predicted error vector as an input to the MPC controller. We can consider future process output as a process state and use modified state space for process modelling. Profit Thick is the model of the thickener with Profit Controller technology. The variables of the model are:
– 7 –
• • •
CV- Controlled variables: Mechanical Torque, Differential Inventory Accumulated Charge, Underflow Solid average percent, and Level Interface DV- Disturbance variable: Pulp Volumetric Flow discharge MV- Manipulated variables: Flocculant dose and % Average Discharge Opening
Figure 5 shows the matrix of relationships proposed for this controller and Figure 6 shows the configuration parameter for the Step Test. MV1: Flocculant dose
MV2: % Avg. Discharge opening
DV1: Pulp volumetric flow discharge
Range
[4-5] grs/Ton
[0 -100] %
[4500-5500]m3/hr
Step size
1 gr/Ton
15%
N/A
Settling Time Max Settling time
Figure 5 Expectations matrix
2,5 hours
Min Settling Time
1 hour
Figure 6 Parameter configuration for the Step Test
Figure 5 can be used to investigate the expected effect on the Output Variables CVs due to the different changes in the Manipulated Variables (MVs): •
Flocculant dose [grs. / Ton]: The expected effect of flocculant is increased sedimentation rate by the group of "floc" that will carry the load compaction zone in a shorter time. The most likely outcome of this effect is a pulp with higher % solids, an increased height of clear water, a higher value of torque.
•
% Average Discharge Opening [%]: The expected effect in Underflow of the thickener is the status of "relief" from the thickener. If the Underflow solid average percent, Mechanical Torque or Differential Inventory Accumulated Charge is too high, the % average discharge opening is sped up to increase the removal rate of slurry from the thickener. If the Underflow solid percent, Differential Inventory Accumulated Charge or Mechanical Torque is too low, the reverse action occurs.
Figure 7 Show the different MVs, CVs, and DVs relation with Profit Thick Strategy.
Figure 7 Profit Thick strategy
– 8 –
RESULTS AND BENEFITS Evaluation conditions
Profit Thick T7 controller assessment was conducted during the months of April and May 2010. The evaluation process consisted in a series of ON/OFF tests. During this period the thickening process operates for four days with Profit Controller and four days without it. Well defined protocols were used to establish the operating conditions for the test, so similar conditions were defined for the operation: Profit Thick ON and Profit Thick OFF. These conditions are: Feed volumetric flow >= 4000 m3/hr • Low Limit Underflow solid average percent > 57 % • High Limit Underflow solid average percent < = 60 % • Flocculant dose >= 4grs/Ton • Operator’s involvement
User involvement was present throughout the implementation process of Profit Thick T7 applications and during all development stages. Constant contact with operation personnel has enabled to satisfy operator needs with the continuous improvement of the applications. This task participation was highly relevant on the results obtained and improved in time, especially on the high application use. Strategy utilisation
Since applications were delivered to operation, they achieved mean utilisation of 96.1%. Their monthly usage in normal months is larger than 95%, as shown Figure 8 and Table 1.
Table 1 Profit Thick usage Month
% usage
Month
Oct-09 Nov-09 Dec-09 Jan-10 Feb-10
99.7 95.3 96.2 96.1 maintenance
Mar-10 Apr-10 May-10 Jun-10 Average usage/month
% usage
92.9 100.0 88.4 100.0 96.1
Figure 8 Usage Profit Thick T7
Stability improvement using Profit Thick
The results obtained with RMPCT respect to manual operation are: Movement of the % discharge opening average has been reduced in 56% in variability as • shown in Figure 9 The Underflow solid average percent has been reduced by 22.3% in variability but a similar • mean value as shown in Figure 10 – 9 –
• •
Mechanical torque has been reduced by 35% in variability and 29.2% in mean value as shown in Figure 11 Level interface has been reduced by 37.57% in variability but with a similar mean value as shown in Figure 12
Figure 9 Histogram of changes in % d ischarge opening Profit OFF vs. Profit ON
Figure 10 Histogram of Underflow solid average percent Profit OFF and Profit ON
Figure 11 Histogram of Mechanical torque in the rake Profit OFF and Profit ON
Figure 12 Histogram of level interface Profit OFF Profit ON
For example, if you increase to 59.5% of solid in the Underflow solid average percent. This could be possible because: •
The Underflow solid average percent can be raised, due to solid improvements in the standard deviation
•
The improvements gained through the reduction of the standard deviation of the mechanical torque and the level interface that were observed during testing.
– 10 –
Increasing from 59% to 59.5% the Underflow Solid average benefits the process by reducing the fresh water consumption in 2.8% per each Thicks High Rate. But this must be confirmed by Plant Test.
CONCLUSIONS AND BENEFITS OBTAINED
Profit Thick has been a successful implementation approach, teaming-up with plant personnel. The strategy acceptance from plant operators is reflected in the large operating availability, shortly achieved. The analysis shows a statistically significant improvement when using Profit Controller in the process of thickening for the values of the standard deviations of Torque and Level Interface. The use of Profit keeps variables process within established ranges. First continuously manipulates valves discharge with smooth movements and ultimately manipulate the flocculant dose, which is very suitable for the treatment of ore in unfavourable conditions. This allows working with a higher average boundary for Underflow Solid average percent because the variability has been reduced, operating with smoother movements average discharge opening valve. It is recommended to implement this control strategy in all the thickeners of the concentrator because the expected benefits are to increase by 0.5% the Underflow Solid average percent in the thickener which represents 2.8% of water used by the Concentrator for processing per day for each Thick.
NOMENCLATURE Q = T x (D1 - D2) Q = Flow of water m 3/day T = Tonnage of tailings thickener Ton/day D1 = Dilution of the pulp in the thickener feed D2 = Dilution of the pulp in the of thickener Dilution = (100 - Underflow Solid average percent) / Underflow Solid average percent Controller OFF
D1 = (100 - 38) / 38 = 1.632
Controller ON
D3 = (100 – 59.5) / 59.5 = 0.6807
D2 = (100 - 59) / 59 = 0.695
Q2 = T x (D1 - D3)
Q1 = T x (D1 - D2)
Q2 = 170000 x (1.632 – 0.681)
Q1 = 170000 x (D1 –D2)
Q2 = 161670 m3/day
Q1 = 159290 m3/day
Therefore, increasing from 59% to 59.5% the Underflow Solid average percent to the same conditions. The flow of water is: Q2 - Q1 = 161670-159290 = 2380 m 3/day. The use of fresh water is 0.5 m 3 per tonnage of ore processed. Profit Thicks saves 0.028 m 3 fresh water per Ton procesed, i.e. 2.8% saves fresh water per each Thicks High Rate .
– 11 –