GB2618995A

GB2618995A - Improvements in or relating to engine control

Info

Publication number: GB2618995A
Application number: GB2205759.0A
Authority: GB
Inventors: Liu Zihao; Haas Oliver; M Herreros Jose; Moradinegade Dizqah Arash; Schaub Joschka
Original assignee: Coventry University
Current assignee: Coventry University
Priority date: 2022-04-20
Filing date: 2022-04-20
Publication date: 2023-11-29
Also published as: GB202205759D0

Abstract

Disclosed is a system and corresponding method for controlling an engine 21. The invention comprises at least two hierarchically connected controllers 22, 23 for performing Model Predictive Control (MPC) of the engine in accordance with one or more performance objectives and one or more MPC constraints. Each controller is aimed at a different dynamic response of the system and has a different prediction horizon, prediction step and control interval. An engine monitoring apparatus 24 is operable to detect engine output variables, the engine monitoring apparatus is connected to the controllers. Each controller is operable to set desired values of engine output variables in line with its performance objectives and monitor the variables in accordance with the MPC constraints and data received from the engine monitoring apparatus, the controllers are arranged such that a higher-level controller is arranged to output data to an immediately subsequent lower-level controller. At least two controllers are activated sequentially from the highest-level controller, and the, or each, lower-level controller is only operable to control the relevant dynamic response of the engine when the immediately previous higher-level controller has completed computing the control actions.

Description

IMPROVEMENTS IN OR RELATING TO ENGINE CONTROL

Technical Field of the Invention

The present invention relates to improved control methods for systems exhibiting fast and slow dynamics to be controlled simultaneously. Particularly, but not exclusively, the invention relates to improvements in aiipath management of internal combustion engines. This is achieved via model predictive control strategies that are capable of using look ahead information, and serves to improve emissions and fuel economy.

Background to the Invention

Whilst modern internal combustion engines (ICEs) are relatively efficient in terms of energy efficiency as compared to other types of internal combustion engines, the emissions from ICEs can often contain significant amounts of Nitrogen Oxide (N0x) and soot. As such, air path management techniques, such as Exhaust Gas Recirculation (EGR) and Variable Nozzle Turbochargers (VNTs), are often applied to ICEs to optimise performance and reduce emissions of the engine. The use of EGR and VNT also affects the fuel economy by reducing the pumping loss, which is the pressure difference between the exhaust and intake manifolds.

EGR is a process which occurs where a portion of the exhaust gases are recirculated back into the engine combustion chamber. EGR is used to reduce NOx emissions and potentially increase energy efficiency by dropping the peak combustion temperature of the engines A modern ICE consists of two or more EGR loops whose flows of recirculated gases are controllable. This helps to change the recirculation rate of exhaust gases with different pressures.

A VNT is another controllable component in modern engines that reuses heat energy of the exhaust gas to rotate a compressor to compress the intake air into the combustion chambers. VNTs consists of one or more variable nozzle turbine(s) that spin a compressor. VNT increases the mass oxygen delivered to the combustion chamber which allows the engine to deliver greater torque.

Air path management for engines is conventionally performed by a tracking 30 type controller which react to the change of engine operating points and track the desired set points of the pressure of intake manifold and oxygen concentration of cylinder Conventional control methods for ICEs comprise multiple controllers which are either 'cascaded' (i.e., one controller calculates the reference point of a lower-level controller), or arranged hierarchically (i.e., each level handles a specific set of states independently or using the given information from the higher levels). The term hierarchical control is used more generically and shall include the applications of cascade control.

MPC is a model-based optimal control approach that uses a model to predict the response of a dynamical system to compute the optimal control actions that minimises an objective function over a given horizon subject to constraints. The horizon is known as the prediction horizon, which spans over a number of control intervals. The control interval is the time duration between two successive control actions.

It is challenging to develop hierarchical MPC strategies to consider fast and slow dynamics where the strategy adopts prediction models with multiple sampling IS rates. This is known as Multi-rate' or 'dual-rate' MPC. A 'multi-rate' MPC, however, presents a significant computational challenge as accommodating both fast and slow dynamics could result in an optimisation problem with a large number of degree of freedoms. For example, a hierarchical air path controller, with a slower sampling rate for the prediction of boost pressure and a faster rate for the prediction of cylinder oxy2en, can be computationally infeasible and may need novel real-time solvers.

Therefore, it is an object of the present invention to at least partially overcome and/or alleviate these issues, through provision of an improved MPC suitable for use in an internal combustion engine.

Summary of the Invention

According to a first aspect of the present invention, there is provided a system for control of an engine, the system comprising: at least two hierarchically connected controllers operable to perform Model Predictive Control (MPC) of the engine in accordance with one or more performance objectives and one or more MPC constraints, each controller being aimed at a different dynamic response of the system and having a different prediction horizon, prediction step and control interval, and an engine monitoring apparatus operable to detect engine output variables, the engine monitoring apparatus being connected to the controllers, wherein each controller is operable to set desired values of engine output variables in line with its performance objectives and monitor said variables in accordance with the MPC constraints and data received from the engine monitoring apparatus, the controllers being arranged such that a higher-level controller is ananged to output data to an immediately subsequent lower-level controller, wherein the at least two controllers are activated sequentially from the highest-level controller, and the or each lower-level controller is only operable to control the relevant dynamic response of the engine when the immediately previous higher-level controller has completed computing the control actions.

The provision of at least two controllers using MPC of the engine allows for complex dynamic problems involving distinct timescales (i.e., problems requiring distinct values for dynamic response, prediction horizon and control interval) to he solved in a manner that reduces computational complexity. The or each controller is operable to separately consider the relevant dynamic problems independently, or having received input from the other controller/s, which consider problems of a different titnescale.

The controllers are formulated to simultaneously consider the minimisation of performance objectives (including NOx, soot emission, fuel economy and the like) whilst having direct control over aspects of the engine's function. The calibration involves the determination of weightings for each performance objective. Compared to known systems that requires the storage of set points which are tracked by lower-level controllers, it also removes the extensive development work to find the appropriate set points.

Each controller computes the optimal control action that minimises the relevant objective function. The present invention incorporates the relevant performance objectives (including emissions, fuel economy etc.) into the relevant objective functions for each controller. This further enables easier calibration of the controllers Additionally, this allows the system of the present invention to function where there is a time delay, as the higher-level controller can accommodate the time delay in its control by considering a longer horizon than the lower-level controllers, which minimises the negative effect of the delay.

Throughout this specification, the phrase lower-level controller' should be understood to include the inner controller (except in circumstances where the relevant controller is referred to as outputting data to a specific inner controller).

The or each lower level controller may activate only when Ihe immediately previous higher-level controller has completed computing the control actions.

The or each lower-level controller may have an equal or faster dynamic response than the immediately higher-level controller.

The or each lower-level controller may have an equal or shorter prediction horizon than the immediately higher-level controller.

The or each lower-level controller may have an equal or shorter control interval than the immediately higher-level controller.

The or each lower-level controller may have an equal or shorter prediction step than the immediately higher-level controller.

The system may be operable to utilize look-ahead information. The look ahead information may comprise future set points or parameters for the control of the engine. The future set points for control may comprise torque demand, engine speed and the like. The look ahead information is provided by an external system, such as an on-board computer. The external system may exploit the current location of the vehicle within a known map. The external system may also exploit information relating to driver intent, such as cruise control set points and driving habits.

In embodiments which utilize look-ahead information, the or each controller may be operable to utilize the look ahead information. Alternatively, and preferably, only the higher-level controller may be operable to utilize look ahead information.

The system may comprise a set of weightings according to the performance objectives and/or MPC constraints. The or each controller may use these weightings while controlling the engine. The weightings may relate to the desired outcomes of the control. For example, the weightings may relate to achieving a desired balance between Nitrogen Oxide (N0x) and soot emissions. The weightings may relate to achieving a desired level of pumping loss of the engine. Preferably, the weightings are set such that the to minimise pumping loss, NOx and soot emissions. The weightings may change dynamically during the use of the engine. The weightings may be set in accordance with a user's inputs.

The engine may be an internal combustion engine. In particular, the engine may be a compression ignition engine. The engine may be equipped with one or more EGR(s) and/or VNT(s). The system may he operable to control said EGR and/or the or each VNT. Each controller may be operable to directly control the EGR and VNT valve positions. The higher-level controller may be operable to control the or each VNT valves. The or each lower-level controller may be operable to control the EGR valves.

Each controller may be operable to solve a different optimal control problem to find optimal valve positions for the EGR and the or each VNT. In particular, each controller may be operable according to different performance objectives and constraints. Each controller may not be operable to set points for the immediately lower-level controller. Preferably, each controller may be operable to set and track set points internally, according to the relevant performance objectives and constraints. This allows each controller to individually optimise the process it is controlling, without external demands set by higher-level controllers.

Each controller may be operable according to a range of values of the prediction /5 horizon. The value of the prediction horizon used by a given controller may be set when the controller is calibrated. The value of the prediction horizon may be within a range of the order of seconds or milliseconds depending on the response time of the characteristic dynamics. In one particular example, the value of the prediction horizon may be anywhere within a range of the order of 0.1 to 10 seconds for the intake manifold pressure dynamics. The decision of prediction horizon considers simultaneously the computational requirement, the duration of system dynamics, and the associated time delays.

Each controller may be operable according to a range of values of response time of a specific dynamics of the system. The value of the dynamic responses used by the controllers may be set during calibration. The system response time may be anywhere within a range of the order of ten of milliseconds to minutes, hours or days.

Each controller may be operable according to a range of values of the control interval. The value of the control interval used by a given controller may be set when the controller is calibrated. The value of the control interval may be anywhere within a range of the order of 5-50 milliseconds. Preferably, the control interval may range may be in the range of 5-10 milliseconds. More preferably, the control interval may be 10 milliseconds. Alternatively, in one particular example, the control interval may be of the order of 20 milliseconds for a passenger car engine.

Each controller may be operable according to a range of values of the prediction step. The value of the prediction step used by a Oven controller may be set when the controller is calibrated. The value of the prediction step used by a given controller may be identical to or greater than the control interval. In one particular example, the value of the prediction step may be anywhere between 0.01 to 0.1 seconds.

The system may be operable to simultaneously balance multiple conflicting 20 performance objectives. In one particular example, the system may be operable to simultaneously balance the emissions levels of NOx, soot emissions, and minimise pumping loss.

The engine monitoring apparatus may be operable to detect operating states that are necessary for the control of airpath. Examples of these operating states include: (present) engine speed, coolant temperature, intake manifold temperature, intake manifold pressure, cylinder oxygen concentration, air intake oxygen concentration, engine-out oxygen concentration, rotary velocity of the turbocharger shaft, exhaust manifold temperature and pressure, engine out NOx concentration, engine out particulate matter concentration.

Each controller may be operable according to an individual prediction model.

According to a second aspect of the present invention there is provided a method of controlling an engine, the method comprising: assigning performance objectives and constraints to at least two controllers, the controllers being operable to perform Model Predictive Control (MPC) of an engine, wherein the controllers are arranged such that a higher-level controller is operable to control a first aspect of the engine's performance and output data to an immediately subsequent lower-level controller, and the or each lower-level controller is only operable to control the relevant dynamic response of the engine when the immediately previous higher-level controller 10 has completed computing the control actions.

According to a third aspect of the present invention there is provided an engine comprising: at least two controllers operable to perform Model Predictive Control (MPC) of the engine in accordance with one or more performance objectives and one or more constraints, each controller being aimed at a different dynamic response of the system and having a different prediction horizon, prediction step and control interval, and an engine monitoring apparatus operable to detect engine output variables, the engine monitoring apparatus being connected to the controllers, wherein each controller is operable to set desired values of engine output variables in line with its performance objectives and monitor said variables in accordance with the MPC constraints and data received from the engine monitoring apparatus, the controllers being arranged such that a higher controller is arranged to output data to an immediately subsequent lower-level controller, wherein the at least two controllers operate sequentially from the higher level controller, and the or each subsequent lower level controller is only operable to control the relevant dynamic response of the engine when the immediately previous higher level controller has completed computing the control actions.

According to a fourth aspect of the present invention there is provided a vehicle comprising the system of the first aspect of the present invention, or alternatively comprising an engine according to the third aspect of the present invention.

The second, third and fourth aspects of the present invention may comprise any of the optional features of the first aspect of the present invention as required or as desired.

Detailed Description of the Invention

In order that the invention may be more clearly understood one or more embodiments thereof will now be described, by way of example only, with reference to 10 the accompanying drawings, of which: Figure 1 shows a known hierarchical control arrangement suitable for MPC of a process.

Figure 2a shows yet another known hierarchical control arrangement suitable for MPC of a process.

Figure 2b shows a generalised example of a hierarchical control arrangement suitable for MPC of a process.

Figure 3 shows an exemplary ICE which is suitable for use in the system of the first aspect of the present invention, or according to the method of the second aspect of the present invention Figure 4 shows a control arrangement suitable for MPC of an engine as part of the system according to a first aspect of the present invention, or according to the method of the second aspect of the present invention.

Figure 5 shows an engine according to a third aspect of the present invention.

Figure 6 shows a vehicle according to a fourth aspect of the present invention.

Figure 7 shows graphs showing the results of using the present invention as opposed to systems known in the art.

Figure 8 shows further graphs showing the results of using the present invention as opposed to systems known in the art.

Figure 9 shows a flow chart of a method according to the method of the second aspect of the present invention.

Turning now to Figure 1, there is provided a schematic diagram of a known cascade control arrangement 1, used to perform model predictive control (MPC) of a process. The arrangement la shows an 'ordinary' cascade ai-rangement, with a pair of controllers 2a,2b arranged such that together the controllers 2a,2b perform MPC of a process 3. One example of such a process 3 is exhaust gas recirculation (EGR) and/or use of variable nozzle turbochargers (VNTs) in an internal combustion engine (ICE) (not shown).

Each controller 2 is operable with the same response time, prediction horizon and control interval. The higher-level controller 2a is operable to set points for the second controller 2b, which is operable to control the process 3 in line with the points set by the higher-level controller 2a. The process 3 is connected to the controllers 2a,2b so as to provide feedback on the state of the process 3, which the controllers 2a,2b IS incorporate into their respective MPC of the process 3.

Figure I also shows a schematic diagram of a process similar to that described above, in this case a 'dual rate' cascade control arrangement. The architecture of the arrangement 1 is identical to that of the above described 'ordinary' cascade control arrangement. except that the controllers 2a,2b now have different dynamic responses, prediction horizons and control intervals. In this case, the higher-level controller 2a has a longer response time, prediction horizon and control interval than the second controller 2b. This allows for the higher-level controller 2a to consider slower dynamics and set points relating to these dynamics for the second controller 2b, which will control the process 3 whilst taking into account faster dynamics.

The 'dual-rate' arrangement shown in Fig. lb functions in the same manner as that shown in Fig la. For completeness, the higher-level controller 2a is operable to set points for the second controller 2b. which is operable to control the process 3 in line with the points set by the higher-level controller 2a. The process 3 is connected to the controllers 2a,2b so as to provide feedback on the state of the process 3, which the controllers 2a,2b incorporate into their respective MPC of the process 3.

Turning now to Figure 2a, there is shown a known hierarchical control arrangement 4a for MPC of a series of processes 5a,5b,...,5n.

Here the 'master' or (highest level) controller 6 is operable to set points for the slave' controllers 7a,7b 7n. Each controller 7a.7b 7n is operable to control the corresponding process 5a,5b,...,5n in line with the points set by the controller 6. The process 5a.5b,...,5n is connected to the controllers 6,7a,7b 7n so as to provide feedback on the state of the relevant process 5a,5b 5n, which the controllers 6,7a,7b,...,7n incorporate into their respective MPC of the process 5a,5b....,5n.

Figure 2b shows another generalised 'n-dimensional' hierarchical control arrangement 4. Here, the highest level controller 6 is operable to control an aspect of engine performance S. and to output data to a lower level controller 7a. The controller 7a is also operable to control a separate aspect of engine performance 5 in line with the data received form the controller 6, and to output data to a lower level controller 7b (not shown). This chain continues, with each controller 7h 7n-1 being operable to control a separate aspect of engine performance 5 in line with the data received form the higher level controller 7a,...,7n-2 and to output data to a lower level controller 7c,...,7n. The lowest level controller 7n is not operable to output data to a lower controller, and simply controls the relevant aspect of the engine performance in line with the data received from the controller 7n-1. The engine performance 5 is connected to the controllers 6,7a 7n so as to provide feedback on the state of the relevant process 5 which the controllers 6,7a 7n incorporate into their respective MPC of the process S. The controllers 6,7a,...,7n are provided with control parameters 8 from external source, such as an onboard computer (not shown).

Shown in Figure 3 is an example of an ICE 10 equipped with EGR (of any /5 known type including single loop or dual loop EGR) 11 and a VNT 12. The ICE 10 is of known type and will be understood by the skilled person.

Figure 4 shows a schematic diagram of a system 20 according to the present invention. The system 20 comprises an engine. In this embodiment, the engine is a ICE 21. The ICE 21 is equipped with EGR and a VNT (not shown). In this example, the dynamics of the problem considered by the higher-level controller 22 (e.g.. boost pressure) are slower than those considered by the second controller 23 (e.g., oxygen concentration).

A pair of controllers 22, 23 are arranged hierarchically, such that the higher-level controller 22 is able to output data to the second controller 23, but not vice versa.

Both controllers 22,23 are arranged such that they can directly control the aspects of the ICE 21 and associated processes. In this embodiment, the higher-level controller 22 is operable to control the position of the VNT valve/s, and the second controller 23 is operable to control the valve/s which regulate the EGR.

Each controller 22,23 is operable to perform MPC of the ICE 21 in accordance 10 with one or more performance objectives and one or more constraints imposed upon to the controllers 22,23.

In this embodiment, the higher-level controller 22 has a prediction horizon of 2 seconds, and a prediction step of 0.1 seconds. In contrast, the second controller 23 has a prediction horizon of 0.1 seconds, and a prediction step of 0.01 seconds. In the embodiment described the control interval for both controllers 22,23 is 0.01 seconds.

hi this embodiment, both controllers 22,23 share the same performance objectives, which are to minimise pumping loss, Nitrogen Oxide (N0x) and soot emissions. The higher-level controller 22 is operable to receive look-ahead information from an on board computer, relating to the future torque demand for the ICE 21.

An engine monitoring apparatus 24 is connected to the ICE 21 and is operable to determine the state of the ICE 21. The apparatus 24 is connected to the first and second controllers 22,23 in order to communicate data to the controllers 22,23 regarding the state of the ICE 21. The controllers 22,23 incorporate this feedback on the ICE 21 state, in order to further optimise the control of ICE 21.

Figure 5 shows an engine 30 suitable for use according to the present invention.

The engine 30 comprises a higher-level controller 31, and a second (and final) controller 32. The higher-level controller 31 is operably connected to VNT 33, which it controls using MPC, according to the performance objectives and constraints assigned to the higher-level controller 31.

The second controller 32 is operably connected to EGR valves 34, which it controls using MPC, according to the performance objectives and constraints assigned to the second controller 32.

The higher-level controller 31 is operable to output data to the second controller 32. In this embodiment, the data comprises the optimal valve position of the VNT 33 which has been computed by the higher-level controller 31. This allows the second controller 32 to control the EGR valves 34 in manner which is cognizant of the state of the VNT 33. Thus, the EGR valves 34 can be controlled in a manner which is complementary to the optimal valve position of the VNT 33, as opposed to being controlled in a noncomplementary manner, which could occur if the second controller 33 did not receive the data on the VNT 33 valve position from the higher-level controller.

An engine monitoring apparatus 35 is connected to the engine 30, the VNT 33 and the EGR valves 35 such that the apparatus 35 can determine the state of the engine 30. The apparatus 35 is connected to the first and second controllers 31,32 in order to communicate data to the controllers 31,32 regarding the state of the engine 30. The controllers 31,32 incorporate this feedback on the engine 30 state, in order to further optimise the control of the VNT 33 and EGR valves 34.

An on-board computer 36 outside of the engine 30 is connected to the higher-level controller 31 in order to provide look-ahead information to the higher-level controller 31. In this embodiment, the look ahead information comprises the future torque demand of the engine 30, which allows the higher-level controller 31 to control the VNT valve 33 in manner allowing the future torque demand to be satisfied, and simultaneously optimising the VNT 33 valve position.

Turning to Figure 6a, there is shown a vehicle 40a comprising a system 41 according to the present invention. The system 41 is a shown in Figure 4, and as discussed in the coffesponding passages of the description.

Turning to Figure 6b, there is shown a vehicle 40b comprising an engine 42 according to the present invention. The engine 42 is a shown in Figure 5, and as 30 discussed in the corresponding passages of the description. An on-board computer 43 (analogous to the computer 36 discussed in relation to Figure 5) is provided inside the vehicle 40b, but external to the engine 42.

The below description of the process is an exemplary process for a compression ignition engine (CIE), also known as diesel engine, which is applicable to each aspect of the present invention, having two controllers. This particular example refers to Figure 4 and adopts the labelling of this figure. The higher-level controller 22 receives the look ahead information from an on board computer (not shown) (not shown) and computes an optimal VNT valve position accordingly. This optimal valve position is transmitted to the second controller 23. The second controller 23 incorporates the optimal valve position into its computation of the optimal EGR valve positions, without using the look ahead knowledge.

hi the embodiment described in Figure 4, the controllers 22,23 both operate according to an economic control problem, as set out below. The subscripts XII and ATL. refer to the quantity X in respect of the first and second controller 22,23 respectively.

IS The subscripts X",," and Xect refer to measured and estimates values of X respectively.

Subscripts (X)(h,/,,./ denote variables related to high pressure EGR, low pressure EGR and VNT respectively. The diagonal matrix T = dingh, t" 14 contains the response time of each valve. t, is the base sampling time, namely 0.01 s. it denotes [fir, in,tkr.

The prediction horizon for variable X is n(X). The vector y = pli,,,]1. contains the output states (for NOx, soot, and pumping loss respectively). Functions, unless with subscripts, are denoted in a generalised manner as f(X). The operating point of the engine is denoted p =[ne, bnzep]', where ne is the rotary speed of thc engine and innep is the brake mean effective pressure. The look-ahead information contains the current and future engine speed and torque, denoted as = [fie, bmep]T The prediction models of the in-cylinder concentration of NOx and soot, pumping loss and the mass of the air inducted into cylinders are as follows: Xnox,k = exp (i (92,k, X0,1c; P)) ( la) Xsoot k = exP (f (P2 kfr X0,k; 1)) Ploss,k = f (P2,k.itk; P) = f (p2,k; P) where the multi-parametric polynomial function f and the corresponding identification method are described in [insert reference]. Here xo is the cylinder oxygen concentration before combustion and x""x and x"", are the cylinder NOx and soot concentration after combustion respectively. The valve dynamics are identical in the situ and in the prediction model, as follows: -tsTuk±i + (1 -tsT)fik (2) Here LI flay) and lifiuvl are the commanded and actual valve positions (of the high pressure EGR, low pressure EGR and VNT respectively). The dynamic states are modelled as in (3) below. The boost pressure (p2) is identified at the sampling rates of 0.1 and 0.01 s for the first and second controller respectively. The prediction model for higher-level controller 22 (3a) does not consider the use of EGR. Instead, the effect of EGR is considered by second controller 23 in (3b).

P2,H,k+1 = fF1032,11,k, X0,11,0 ± ft (Üv,k; p) (3a) P2,L,k+1 = IL(92,110 X 0,k) fL(llk; P) (3b) = IL(132,1,k, X 0,k) IL(fik; (3c) Finally A0 is the fraction of Oxygen Fuel Ratio (OFR) over its stoichiometric value: Oncyl,kX0,k, -air In fuel0FR sto ich Here ilithr and M02 are the molar mass of air and oxygen respectively, and m"),/ is the cylinder mass intake, and tn.*/ is the mass fuel injection per stroke. Equivalently, n1:fin' may also be expressed in the form of fuel injection quantity per cylinder, or for all cylinders. The Optimal Control Problem (OCP) of the higher-level controller 22 and its stage cost function /Hi are shown in (5) and (6a), respectively. ;minimise (5a) (u,,xo,H)EUvxXxo Ellin 1H 0/ v,i; s.t. (2), (1), (3a), (4) (5h) (1d) AO,k (4a) P2,0 = P2,mea, x0,0 = x0,est = ftv,mect, TO) = [ne,mect, inneP est] T (5d) unnin uv 5 uvamax (Se) xvzin x xmax (50 Ao AO,min (5g) k E {0,1, ...,nw -1} where: := a x f(Nyyj QH(Nyyi) + ww,i(Nujuv,i uv,eff))2 (6a) 14/11,2(NAuvAuvi32J = (611) tr(2H)+ EL, wHf Hereltvieff is the most efficient position for the VNT valve (this is variable and the actual value is operating point specific), and N is the prediction horizon for the variable y. The quantities w are the weightings of each objective function term. ;Equations (7) and (8a), represent the OCP and the stage cost function /Hi of the second controller 23 respectively. ;minimise Vail, (,,, lY 15) (7a) s.t. (2), (1), (3b), (3c), (4) (7b) P2,0 = P2,mect, X0,0 = x0,est (7C) 1111,0= fitt,mea, 111,0 = Ui,mea (7c1) U1, = U0,H (7e) uh,mth. up Uh,max* ulyttin U0 UI,max (70 xmin 5 x xin" (7g) Au AO,min (7h) k E -1} where: : = ie X { (Aly Q (NyYi) WL,1 (Nuh U71,02 + wL,2(Nutu1,02 +144,3(kuvauh,32 + wL,4(NAutaut,321 fl - 1 (811) tr(Q 0+ EL The stage cost functions penalise the variation of the control signals and the energy of valve positions which is expressed as the quadratic values of the control valve positions. VNT is exceptionally penalised in (6a) with respect to its most efficient position u,"eff. The value of u","ff is operating point dependent, and the penalisation encourages the VNT to operate away from choking and surging boundaries.

The higher-level controller 22 computes the optimal trajectory of VNT valve position and oxygen concentration. The second controller 23 receives the u(1) (denoted uv' m in (7e)) and assumes that it stays constant for the prediction horizon.

Meanwhile, the optimal trajectory of oxygen concentration is discarded.

Both OCPs exploit the limit ono as the driveability constraint. The satisfaction of (5g) and (7h) allows sufficient fuel to deliver the demanded torque. The 20 limits are enforced as soft constraints that allow minor violations. This ensures that the OCPs are feasible during rapid increase of torque demand.

Solving (5) and (7) requires the first and second controllers 22,23 to agree on an optimal trade-off between boost pressure and in-cylinder oxygen. Due to the emission trade-off, the 'NOx favoured' higher-level controller 22 will work against the 'soot-favoured' second controller 23, causing systematic degradation of performance. It is found a similar weighting of Qh and QL is sufficient to prevent the conflicting emission objectives. This is because the emission objectives are less sensitive to the change of boost pressure than that of cylinder oxygen, which is governed by the higher-level controller 22.

The normalisation factors Nx are of values depending on the range of signals and the engine, whilst a and //normalise weightings to form a convex combination of objectives in (6a) and (8a).

Table 1 summarises the emission, fuel consumption and torque tracking results of the engine with the proposed controller as compared to an optimal controller that does not take look-ahead information into account (i.e.. MPC w/o lookahcad information), as well as to a production-line controller. Table 1 excludes the data during engine idling due to possible start-stop strategy and to emphasise performance variation during engine transients. Table 1 shows data from the Heavy (1-1) and Extra Heavy (EH) cycles of the Worldwide harmonised Light vehicles Test Cycles (WLTC), taken from a desktop simulation running the method disclosed above, and a Hardware-in-Loop (HIL) test.

Table 1

WLTC Stage Improvement (%) against Improvement (%) aminst eMPC Production-Line Controller vv/o Look-ahead Info NOx Mass Soot Torque NOx Mass Soot Torque (g) Mass Tracking (g) Mass Tracking (g) RMSE (g) RMSE ((Nm)-1) ((Nm)-1) Desktop Simulation H 7.2 9.4 41.1 4.1 5.5 10.6 EH 13 15.1 28.7 13.4 13.4 20.8 HIL Implementation H 9.8 143 -63 5.2 6.1 6.6 EH 18 153 1.4 4.4 6.8 23.7 Compared to the production-line controller the use of lookahcad information enabled by the control system of the present invention improves the performance of engine in every aspect. As compared to the MPC w/o look-ahead information, the control system of the present invention allows a greater reduction of soot. The control system of the present invention also improves torque tracking steadily as the duty cycle becomes heavier. This is because the control system of the present invention is able to use the lookahead information to prepare ahead for larger load variance in heavier duty cycles.

Fig. 8 shows a set of selected transients from the 'heavy' and 'extra-heavy' parts of the Worldwide harmonised Light vehicles Test Cycles (WLTC). Upon each tip-out event, the system of the present invention achieves up to 5% higher net indicated efficiency by decreasing the boost pressure earlier (e.g., at 1159 s, 1162 s) than the other controllers, which perform more conservatively. During rapid fall-and rise of the torque demand at 1545 s, the developed controller maintains the boost pressure and allows more air intake than the other controllers. As a result, the engine delivers more torque when the demanded torque increases rapidly. The look-ahead information also allows the system of the present invention to build up the boost pressure more economically -that is, the provision of boost pressure is just sufficient for delivering the required amount of torque. Between 1155 and 1172 seconds, all controllers have a similar buildup of boost pressure initially, and the system of the present invention relaxes the boost earlier without compromising torque tracking. Also, the system of the present invention shows consistently lower peaks of boost pressure whilst maintaining competitive torque tracking. The proactive relaxation of boost pressure allows the engine to achieve higher indicated efficiency when possible. This reflects in the fuel economy advantage of using the system of the present invention. In terms of emission, the production-line controller shows an undershoot tendency in oxygen concentration which causes soot spikes, whilst the systems that utilise MPC use EGR to a lesser extent. Both the system of the present invention and the system which uses MPC w/o look-ahead information have kept (A0)-1 under the relevant threshold. In contrast, the production-line controller had minor violations of this threshold, since the injection controller cuts fuel reactively using an estimation of the cylinder mass and oxygen concentration.

A predictive controller can benefit from the sharp sequences of falls and rises of torque by gear-shifting to particularly improve the NOx emission and torque tracking. However, automatic gear-boxes or the fast gear-shifting by drivers may reduce this benefit. The performance of the system of the present invention is additionally verified using a WLTC data set of automatic transmission with start-stop strategy (see Fig 9). The system of the present invention demonstrates consistent improvement of 7.2%, 0.1%, 8.3% and 9.8% in torque tracking, fuel economy. NOx, and soot emissions, compared with the system which uses MPC w/o look-ahead information.

Figure 9 shows a flowchart of an exemplary method for controlling an engine. At step 101, the engine being controlled is turned on. At step 102, a driver input (such as a demand for specific power output via use of the accelerator pedal) is received by the engine. After the driver input is received by the engine, look-ahead information is provided to the high level controller at step 103.

Engine feedback is acquired by the controllers (from the engine monitoring apparatus) at step 104, and the controllers use this feedback to estimate the operating state of the engine at step 105.

Once the engine state has been estimated at step 105, the higher level controller determines the determine control action at step 106. Specifically, at step 106, the higher level controller determines the required control action by solving the system of equations (5a-5g). Once the control actions have been determined, these are communicated to the immediately subsequent lower level controller at step 107.

The lower level controller then calculates the required control action at step 108.

Specifically, at step 108, the lower level controller cleteimines the required control action by solving the system of equations (7a-7h). Finally, at step 109, the control actions determined at steps 106 and 108 are communicated to the engine, which is controlled in line with the computed control actions. The above steps 102 through 108 are repeated as necessary, according to each new driver input.

It will be understood that the flowchart of Figure 9 can equally apply to generalised 'n-dimensional' systems, such as that shown in Figure 2b. For example, using the numbering of Fig. 2b, after step 108 where the lower level controller 7a determines the control action, then the lower level controller 7a may communicate these to a subsequent lower level controller 7b, as in step 107. This 'looping' process occurs until the final, lowest level controller 7n has computed the control actions, at which point there are no subsequent lower level controller, and the flowchart proceeds to step 109.

The one or more embodiments are described above by way of example only. Many variations are possible without departing from the scope of protection afforded 30 by the appended claims.

Claims

CLAIMS1. A system for control of an engine, the system comprising: at least two hierarchically connected controllers operable to perform Model Predictive Control (MPC) of the engine in accordance with one or more performance objectives and one or more MPC constraints, each controller being aimed at a different dynamic response of the system and having a different prediction horizon, prediction step and control interval, and an engine monitoring apparatus operable to detect engine output variables, the engine monitoring apparatus being connected to the controllers, wherein each controller is operable to set desired values of engine output variables in line with its performance objectives and monitor said variables in accordance with the MPC constraints and data received from the engine monitoring apparatus, the controllers being arranged such that a higher-level controller is arranged to output data to an immediately subsequent lower-level controller, wherein the at least two controllers are activated sequentially from the highest-level controller, and the or each lower-level controller is only operable to control the relevant dynamic response of the engine when the immediately previous higher-level controller has completed computing the control actions.
2. A system as claimed in claim 1 wherein the or each lower-level controller has an equal or shorter prediction horizon than the immediately higher-level controller.
3. A system as claimed in either claim I or claim 2 wherein the or each lower-level controller has an equal or shorter control interval than the immediately higher-level controller.
4. A system as claimed in any preceding claim wherein the or each lower-level controller has an equal or shorter prediction step than the immediately higher-level controller.
5. A system as claimed in any preceding claim wherein the system is operable to utilize look-ahead information.
6. A system as claimed in claim 5 wherein the look ahead information comprises future engine output requirements.
7. A system as claimed in claim 6 wherein the future engine output requirements comprise torque demand and engine speed.
8. A system as claimed in any of claims 5 to 7 wherein the or each controller is operable to utilize the look ahead information.
9. A system as claimed in any of claims 5 to 7 wherein only the highest-level controller is operable to utilize the look ahead information.
10. A system as claimed in any preceding claim wherein the engine is an internal combustion engine (ICE).
11. A system as claimed in any preceding claim wherein the engine is a compression ignition engine (CIE).
12. A system as claimed in any preceding claim wherein the engine is equipped with exhaust gas recirculation (EGR) and/or one or more variable nozzle turbochargers (V NTs).
13. A system as claimed in claim 11 wherein each controller is operable to directly control the EGR and VNT valve positions.
14. A system as claimed in claim 12 wherein the higher-level controller is operable to control the VNT valves, and the or each lower-level controller is operable to control the EGR valves.
15. A system as claimed in any preceding claim wherein each controller is operable according to a different optimal control problem.
16. A system as claimed in any preceding claim wherein each controller solves an optimisation problem to find optimal valve positions for the EGR and the or each VNT.
17. A system as claimed in any preceding claim wherein each controller is not operable to set points for the or each lower level controller.
18. A system as claimed in any preceding claim wherein each controller is operable to set and track points internally, according to the relevant performance objectives and constraints.
19. A system as claimed in any preceding claim wherein the system is operable to simultaneously balance the emissions levels of Nitrogen Oxide(N0x) and soot.
20. A system as claimed in any preceding claim wherein the system is operable to minimise pumping loss. NOx and soot emissions.
21. A method of controlling an engine, the method comprising: assigning performance objectives and constraints to at least two controllers, the controllers being operable to perform Model Predictive Control (MPC) of an engine, wherein the controllers are arranged such that a higher-level controller is operable to control a first aspect of the engine's performance and output data to an immediately subsequent lower-level controller, wherein the or each lower-level controller is only operable to control the relevant dynamic response of the engine when the immediately higher-level controller has completed computing the control actions.
22. An engine comprising: at least two controllers operable to perform Model Predictive Control (MPC) of the engine in accordance with one or more performance objectives and one or more constraints, each controller being aimed at a different dynamic response of the system and having a different prediction horizon, prediction step and control interval, and an engine monitoring apparatus operable to detect engine output variables, the engine monitoring apparatus being connected to the controllers, wherein each controller is operable to set desired values of engine output variables in line with its performance objectives and monitor said variables in accordance with the MPC constraints and data received from the engine monitoring apparatus, the controllers being arranged such that a higher controller is arranged to output data to an immediately subsequent lower-level controller, wherein the at least two controllers operate sequentially from the higher level controller, and the or each subsequent lower level controller is only operable to control the relevant dynamic response of the engine when the immediately previous higher level controller has completed computing the control actions.
23. An engine as claimed in claim 23 comprising any of the features of claims 1 to 19.
24. A vehicle comprising: a system according to any of claims 1 to 20, or an engine according to either claim 22 or 23.