A NEW APPROACH TO THE SOLIDIFICATION MODELING OF CASTING PROCESSES Dr. V. S. Lerner, University of California at Los Angeles, UCLA, CA, USA Dr. Y. S. Lerner, University of Northern Iowa, Cedar Falls, Iowa, USA ABSTRACT A new systemic mathematical - informational approach based on Informational Macrodynamics (IMD) is introduced for solidification modeling and optimization of the casting processes. The IMD model takes into account the interrelated thermal, diffusion, kinetic, hydrodynamic, and mechanical effects essential for the given casting process. The optimum technological process parameters are found by simultaneous solution of problems of identification and optimal control based on the extremum of the entropy functional, evaluating a generalized assessment of the physicochemical properties of castings. For the physical system considered, the IMD structures of the optimal model are connected with controllable equations of Nonequilibrium Thermodynamics. This approach was applied to horizontal continuous casting of ductile iron and results were compared with experimental data. These comparisons have validated the accuracy and usefulness of the new approach. INTRODUCTION Traditional numerical modeling of the casting process is using mathematical models and approaches that mostly describe a separated process's phenomena, or severely cross phenomena, which simulate mold filling and solidification (1). These models need massive computations, and are not suitable for operative process control. The proposed method is based on Informational Macrodynamics (IMD) and is used here for a system informational solidification modeling and optimization (2-4) of the casting process taking as an example horizontal continuous casting (HCC). HCC is a relatively new, but promising method of producing near net shape high quality cast products in ferrous materials (gray, ductile and NiResist irons and steel), as well as non-ferrous (aluminum and copper) alloys. In HCC, liquid metal from the transfer ladle is poured into the metal receiver. A water-cooled graphite or copper die is attached to the side of the receiver and bar is pulled out by an extraction system, which controls stroke length and frequency. A special mechanism cuts and breaks the bars to required lengths. The major advantage of this process is a high casting yield of 92-95%, since it eliminates traditional feeder needs. Liquid metal in the receiver plays the role of a preheated riser that continuously supplies liquid metal to feed the bar during solidification. By maintaining an adequate balance between the metal chemistry, temperature, level in the receiver, and drawing and cooling parameters, it is possible to produce defect free high quality continuously cast bars (5,6). THE BASIC IMD CONCEPTS The IMD utilizes Lagrange's Mechanics of Uncertainly, Nonequilibrium Thermodynamics (NT), Systems Information Theory and Probabilistic prognosis, uses a common information language for computer modeling during a combined object's observation, control, and 365 simulation. The IMD informational language represents the model's revealed uncertainty directly via a corresponding information measure. That unifies and integrates different physical models within IMD. IMD contains a formal mathematical model to describe the transformation of random information processes (at microlevel) into the system dynamic processes (at macrolevel). The observed process at microlevel is represented by a set of random interactions connected via a Marcovian chain with a path integral evaluation. This functional accumulates information contributions from the local functionals of the interacting microlevel processes as a collective (macro) functional. Transformation from microlevel to macrolevel is performed by using the informational form of variation principle (VP). The dynamic macromodel follows from the solution of the VP minimax problem for information form of the path functional (2-3). Microlevel stochastics transfer the irreversible macrodynamics that extract order from microlevel randomness. Macromovement occurs along segments of the initial n-dimensional extremals of the functional. These segments are successively joining at discrete points (DP) effectively shortening the initial dimension and ultimately leading to renovation of dynamic process. The main formal result is the LagrangeHamilton's equations for the conjugate informational macrocoordinates, which together with the equation of the differential constraint (DC) describe the model's macrodynamic regularities. DC reflects a ‘deterministic impact’ of microlevel stochastics on the Lagrange Hamilton's informational mechanics of uncertainties. In a chain of superimposing processes, each subsequent process controls one or more the following chain's processes with a possibility of changing the operator. In the IMD-NT macroequation: n dxi Li X i ui , ui Lij X j , Lii 1/ 2rii , rii M [ xi2 ], i j (1) dt j 1 chain of superimposing processes ( i -1, i , i +1) is connected by mutual cross phenomena u, which perform the control function depending on the state coordinate at ( i -1) DP interval: dx 1 ui 2i xi (t ' ) 2 i =2 xi 1 (t ' ) , Lii rii i , vi (t ' ) = 2 xi (t ' ) . (2) dt The applying of needle controls: v(t ' , l ) 2 x(t ' , l ) 2 x(t 'o, l ) , v(l ' , t ) 2 x(l ' , t ) 2 x(l 'o, t ) (3) changes only the sign of the operator components without changing their absolute values. The extremals of the entropy functional are the informational analogies of the solution for the irreversible NT equations with the Onsager condition on the each of extremal segment. The kinetic operator is changing by jump at each DP time-space point (t', l' ): x x L ( x(t ' , l ' )) 2b( x(t ' , l ' )) M x (t , l ' ) [ x(t , l ) (t , l )*] M x (t ' , l ) [ (t , l ) x(t , l )*] (4) t t of the discrete control forming, where the following equality is true Ii I (t j ' , l ' j ) g i (t j ' , l ' j ) g k (t j ' , l ' j ) k (t j ' , l ' j ) (5) Xi Xk and g i , g k are the subsequent equalized components of the generalized transient conductivity. The needle controls select them based on condition of the model controllability. The 1 x generalized forces X= (2b) are formed by the stochastics and applied macrocontrols, t which physically have a quantum (portion) character at the macrolevel interactions. The 366 points of compensation diffusion and kinetics hold the chain connection. Sequence of the chain dependable n-controllable conductivity's components can be reduced to one controllable conductivity, for example, to electrical conductivity, which measuring is most simple. The macrodynamic process is characterized by the discrete extremal intervals, selected from Hamilton's solutions via DC with the time interval discretizations (DP), determined by VP invariants. Hamilton's equations determine, in general, the reversible dynamic solutions, and the DC equation (at the DP's) imposes the dynamic connection with microlevel's stochastics. The DC connects macromodel's operator to statistical characteristics of the random process, in general, in the form of the nonlinear correlations, which, in part, lead to the precise and simple identification equations (2,3). The DC changes the structure and value of dynamic macromodel operator at DPs, creating possibility of modeling jumpwise object's phenomena, connected with abrupt discontinuing or non-smoothness of identified process. The character of specific object's stochastic data imposes the constraints on the macrolevel's variation problem. Mathematically the constraints are determined by the equations of Marcovian stochastics at microlevel, specifically by limitations on the structure of shift and diffusions matrixes of Marcovian stochastic equation, or their widely used nonMarcovian approximations. The information functional integrates these equations into the dynamic constraint at the solution of considered minimax optimization problem (VP). The VP solution uses Lagrange's method of eliminating constraints and Pontryagin's maximum principle. The optimal control synthesis is solved using the corresponding Boltz's problem. Mathematical condition on the constraints and the cost function is expressed by the information analogue of function of action, which satisfies a joint solution both the HamiltonJacobi's and Kolmogorov's equations on some selected set of the extremal's field. This set defines a natural boundary of the variation problem, and later on is selected constructively by the synthesized optimal control functions (2). The applied variation techniques combine the traditional ones with Kolmogorov's equations, which model the microlevel’s Marcrovian stochastics and create a very specific variation minimax problem (2). The informational dynamics and geometry of the macromodel are portaying both uncertainties (4). Information Geometry (IG) visualizes the bound model's hierarchical information structure, space-time locations, configurations, and shapes of interacting dynamic information flows as the systemic categories. The IMD modeling selects the most informative data at DPs responsible for the process phenomena. IMD combines object's observation and prognosis on the basis of chosen performance criteria. Because of possibility of nonpredictable changing of object's characteristics, the operations of object's identification and optimal control are joining in time. It means, both object's identification and optimization problems can be solved simultaneously by applying the same control strategy directed toward minimization uncertainty between current observations and expected prognosis. Minimizing uncertainty requires applying control that extracts maximum information compensating the uncertainty. The successive joining of the piece-segments of model's extremals leads to a sequential consolidation of object's processes (by shortening the initial dimension) and revealing the object's hierarchical systemic structure during the optimal motion. The sequenced hierarchical connections create the hierarchical information dynamic network (IN) of macronodes (Figure 1), formed by informational spectrum of macromodel's operator hio with the optimal path through this network, identifiable in process of the object observation. Restoration of the hierarchy of the identified IN's nodes brings a quantity measure of the level of systemic organization, evaluated by the function of informational macrosystemic dynamic complexity 367 (MC). The structurally stable macrodynamics (with a possible local instabilities) can create the IN's macronodes formed by a unification of collective chaotic attractors. Chaotic regions can generate the second level of stochastic behavior with a possibility of all kinds of chaotic dynamic phenomena. Depending on the identified object's data, the model can reflect the object's singularities and/or nonlinear behavior. The IMD model has 4-levels of hierarchy: 1) statistical microlevel; 2) quantum dynamic level (at the DP locality); 3) dynamic (classical) macrolevel as a result of selection of the initial macrostates and macrotrajectories; and, 4) hierarchical informational dynamic network of macrostructures. This many-levels structure represents a basic analytical model's form, which is concurrently updating, validating, and progressively refined. This structure can capture the essential object's regularities and ignore the irrelevant aspects, with different level of abstractions, including the automatic modeling both sooth and jump- wise phenomena. A combination of the identification, optimization, and disclosure object's hierarchy reveals the object's systemic functions and regularities in process of functioning. It leads to constructive engineering solutions. hio Figure 1. Structure of the space-time Informational Network of the superimposing processes, defined by the initial information spectrum hio , which are cooperating along the hyperbola. The IMD specifics allow us modeling the object's complex phenomena characterized by a variety of superimposition processes. The informational IN network of superimposing processes models the chain of their sequential cooperations by the speed and space's interactions. Both the IMD and NT equations are used to connect the traditional physical approach with IMD's modeling formalism. The virtual INs can be used for mutual 368 communications with possibility of connecting to WEB-based modeling. This communication can involve human activities in decision-making and the corresponding model's adjustments using artificial intelligence approach. Using the IN's formal logic and its information language for description the IMD model and IG operations (4), brings unification for IMD algorithms with an opportunity of formal application to wide diversity of complex objects. The formal set of possible IMD models are classified in terms of their MC-complexity, which depends on the basic identifiable model's parameters. The identification of these basic object's parameters leads to choosing the particular information model's computational code depending on the object's complexity (that includes the size of object's computation). This approach is directed toward the solution of scalability problem for information model with decreasing its cost. Basic IMD information computation software is simpler comparing with traditional massive numerical computations, and can work on regular PC with Java language code. Unified IMD approach brings standardization of module's functions with no software changes and no degradation in efficiency. THE HCC SOLIDIFICATION OPTIMAL MODEL Solidification is created by the HCC’s system of superimposing processes is shown on Table 1, where phase coordinates ( xnt k , xnl k ) is presented as the derivatives and integrals of corresponding ( n k 1) coordinates, associated with physical quantities: temperature , rate of solidification vc , concentration C, rate of mass transfer v m , thermomechanical stresses , strains , density , and pressure p . The number of total variables n is previously unknown. The thermoconductivity (T) is the basic phenomenon generating thermostresses (TS), solidification (S), the mass transfer M, initiated by effective diffusion (D), the phase transformation (PT), the motion of the liquid phase transfer under the influence of the hydrodynamic forces HT. Both the IMD and NT equations are used to connect the traditional physical approach with IMD modeling formalism. For example, to describe heat conduction (T) and solidification (S) the Fourier and Stefan equations are used. Diffusion (D) is described by equations of Fick's first and second laws, associated with the laws of conservation. The equations of thermomechanical stresses (TS) are determined by the bar temperature distribution in time and space, e.g., by solution of equations T, and the equations, which describe the phase conversions (PT), associated with the change in concentration of carbon C= C-C1 in the liquid (C) and solid (C1) phases. To describe the hydrodynamics (H) the Navier-Stokes equation is used, etc. As an optimization criterion for the systemic macromodel, the entropy functional of quality is employed, the minimal value of which determines the maximum order of the bar’s structure. This criterion, determined by the conditions of the extremum of nonequilibrium entropy functional for the process (Table), emerges as a general indicator of bar structural uniformity on the micro- and macrolevels. The order on the microlevel implies structural uniformity of the metallic matrix through the cross section of the bar with uniform distribution of graphite nodules, and the uniform distribution of chemical elements through the cross section of the grain boundaries without segregation. The order on the macrolevel corresponds to production of a bar with a uniform macrostructure without internal defects, with fulfillment of the conditions of directional solidification and compensation for shrinkage. As a result of the optimization problem solution, the regularity of change in the kinetic operator is characterized by 369 successive reduction with time in its intrinsic values, which are equated at specified DP points. The corresponding dynamic processes illustrates the connections between T, S, D, PT processes in the two-phase zone. The ranged processes in optimal model are characterized by three basic parameters: the number of equations n, the variability , and the space k parameters, which are common for the macromodel. Table 1. Scheme of the main interrelated physical phenomena Physical phenomena and their notations Phase-space coordinates l x Phase-time coordinates xt xnl 1 2. Heat conductivity (T) (initial phenomenon) xnl 3. Solidification (S) under temperature gradient xnl 1 (l) ~ l (t) ~ ~ vc l xnt 1 ~ C t xnt 2 2 C ~ t 2 t xnl 2 2 vc ~ l 2 l 5. Mass transfer (M) xnl 3 2 C ~ vm l2 6. Phase transformation (PT) causing stresses and pressure ~ ~ l t t vm 3 l ~ 3 xn 4 l t 8. Hydrodynamic mechanism (H) developing a pressure l xnt 4. Effective diffusion (D) under heat flow 7. Hydrodynamic transformations in liquidsolid state (HT) xnt 1 (t) dl ~ (l) 1. Thermomechanical stresses (TS) xnl 5 vm ~ p(l) l2 xnt 3 t n 4 x 2 C t2 vm 2 ~ pdl ~ 2 t t xnt 5 p(t) xnt 6 p t 2 p l xnl 6 The solution of optimization problem ensures a maximum value of the functional of microand macrostructural ordering. 370 Figure 2 illustrates the diagram of computation of the IMD optimal model's process x(t) xi (t,l,u(ti )) using a given space image with distributed information S per cross-section F . i (t i ) S S F F x[t,l,u(t)] n k invar li i (t i ) x i(t) ti ti v(t i) u (t i ) u (t i ) i (t i ) x i(t) Figure 2. The diagram illustrating the sequence of the IMD optimal model's process computations. The algorithm calculates the models' basic parameters ( n , , k ), which define the model's invariants INVAR, time ti and space li discrete intervals, eigenvalues i (ti ) of differential model's operator, MC-function, speeds CT ,C , simulates the inner vi (ti ) and output ui (ti ) optimal controls. The model’s specific Hamiltonian hv e is used to calculate the optimal hydrodynamics and basic three parameters of new model. OPTIMAL HYDRODYNAMIC CONDITIONS FOR THE IMD MODEL. CALCULATION THE MODEL’S PARAMETERS The systemic solidification macromodel (Table) includes the hydrodynamic components (H, HT) that interact with motion of the liquid and solid phases participating in HCC. Maximal macrostructure order and obtaining a homogenous bar without defects can be reached by minimizing hv e that defines the optimal function of the entropy production = (t , l ) . The problem consists of finding the optimal functions for hydrodynamic variables: the pressure p , the velocity of the liquid phase movement, which are defined by the given function = (1, ) . Let us consider the movement of a non-compressed Newton's liquid within a cylindrical channel, described by the equations: t zo p = , z , 2 (6) r o z ro with the boundary conditions: r (0, R ) ( R ) , R , ( ,1) ( ) =0; =-2g( ). (7) R ro The tangent tension on the liquid-solid phase boundary is given by the equation: f a1 , a ~ , . (8) R 1 371 The equations are written in relative (non dimensional variables) as the time , the radius R and the one of space coordinates l z , and with the absolute variable: t as a time, , as the kinematic and dynamic viscosities, r , ro are the current and fixed channel radiuses accordingly, zo is the axis coordinate, is the density. The entropy production is connected with the tangent tension f by the relation: ( R, ) a 2 f 2 , a 2 ~ (9) where it is assumed that the parameters a , a1 , , are given and fixed. The entropy production for the optimal systemic model is defined by the equation: (1, ) b 2 ot (1, ) (10) where ot (l , ) = t ( ) is the given function of time ( in the IMD equations) , and b is the constant coefficient. Function g ( ) (7) can be expressed via using (10) in the form: b b2 t = a 2 f 2 = a 2 a12 ( ) 2 , g (11) ( t )1/ 2 . R 2a1 a p f at the given t , we obtain By solving the equations (6-9) for an arbitrary function z the following integral equation with regard to f: b f ( ) exp(1 2k ( )) d = C ( t )1 / 2 , C (12) a k 1 0 t 2 where k are the eigenfunctions of equation (6). The solution gets representation: f f , k k f k ( ) exp( 2k ( )) d = 0 C ( t ) 1 / 2 . k 2 (13) After substituting (13) into (6-9), the field of velocities ( R, ) ) can be found. Let us consider the concrete results. Suppose is given by the following equations: = o , = ( o , o , z), o = o ( r 2 ) 1 , o = o ( o r 2o ) 1 , o = o ( ) (14) where o , o = o are determined by the parameter of the optimal model at given ro and . Then the sought function f ' acquires the form: | | r 2 (t ) 1/ 2 1/ 2 25.1 ( o o - [ (0)] ) (15) 1 / 2 + [ ( t )] 4 25.1 [ (t )] From that at (0) =0, (t ) = (t), o =-2.29, o =-1.1456, =0.5 we get the solution 3.3 (t ) (t ) o 25.1 ( ). (16) [ (t )]1 / 2 The pressure changing should be applied to the solidified iron bars by alliterating the metal level within the iron receiver. In the casting process with applied pressure, for fulfillment the directional solidification, the external pressure as a control function should satisfy the relation for o . The feeding and the solidification processes in systemic model are mutually 372 interconnected such a way that the speed of the liquid metal movement is coordinated with changing density of the solidified bar. The found optimal impulse feeding law C ( ln )=C [ ln ( t n , l T , t c , t )] satisfies these conditions, where t n is the starting moment of changing linear feeding velocity C (t); t k is the fixed time interval, t is the current time interval, l T is the space interval of the impulse feeding controls; t c is the time interval of stopping the feeding impulses; Cl is the average solidification speeds. The linear bar size increment satisfies the function l n (C C l )t n 1 , where the component t l C T (1 c ) , t = t k - t n characterizes the average speed of feeding. The difference ( ln ) t tk defines the space distribution of metal that is necessary to compensate a volumetric contraction at l l n , and during the time interval t n 1 . At the real conditions of the gravitational feeding, the impulse cycle takes place, when the continuous speed feeding Cl is equal to the average solidification speed, and the bar high is equal ln . The time of solid skin formation, required to start bar’s extraction or draw, is governed by the conditions needed to control the elastic deformations with compensation for shrinkage. The optimal drawing speed changes in a discontinuous pattern and is characterized by a cycle frequency: drawing-pause. The implemented algorithm allows the computation of the optimal value of casting speed, the drawing and pause intervals, the cast iron temperature in the receiver, the cast bar temperature, the variation in the flow rate and water temperature in the die, and the parameters of die design and secondary cooling conditions. CALCULATION OF CHEMICAL COMPOSITION AND MICROSTRUCTURE PREDICTION Each cross section of continuously cast bar is characterized by its own solidification pipe with the zone of effective diffusion. The concentration of chemical components in this zone depends on the solidification time. With increasing cross section of the bar, solidification time and the time of effective diffusion increase resulting in structural microheterogeneity. To ensure the same concentration of chemical elements in the model bar, as in the actual bar, the model controls the concentration gradient at the boundary of the solidification zone and its distribution within this zone. The problem is solved by selecting the chemical composition of the bar as a function of the initial distribution coefficient. Knowing the change in concentration of chemical elements during solidification of the base bar, the optimal chemical composition of the given bar may compute. The model also calculates amount of alloying elements, for example Sn, Cu, Mo, which are required under real solidification conditions to control the initial distribution coefficient within the zone of effective diffusion and as-cast microstructure. The model calculates a phase transition in the presence of k points of nonequilibrium, particularly, the nodule count (NC), when (n-4+k), or two (n-2+k) elements are precipitated. The radius R of the bar corresponds to the model length of the solidification pipe, which is characterized (at the fixed angle at its vertex) by the segment of the IN’s helix of the cone l (n+1)), where the number of elements 2 2 (Ln+1/Ln+k-1) 2(n-2-k)(l+ ) (17) 373 is found. NC per unit of cross-sectional area S= R , precipitated at the k-th point, is: NC=4 /S (n-2+k) 2 (Ln+1/Ln+k-1) 2 (l+ 2 ) 2 (18) where n and are the models' basic parameters, L(n+1)=L(n, ) is the model length of the solidification cone; Ln+k-1=L (n, ) is the model length of the effective diffusion zone calculated on computer. In the optimal model, graphitization occurs in the spatial interval L (n-4) with the maximum NC (L (n-4)) nodule count. The computerized method calculates the NC at any point along the entire cross section of the bar with a sufficient accuracy (7). 2 CONCLUSIONS A new information approach to the simulation and control of the solidification process using the IMD has been described. This new computation techniques and software have been applied to the horizontal continuous casting of ductile iron. Comparisons with the results of actual test runs have validated the accuracy and usefulness of this new approach. ACKNOWLEDGMENTS The authors would like to thank Professor J.T. Berry of Mississippi State University for helpful suggestions rendered during the preparation of this paper. REFERENCES 1. Godsell B: ‘Continuous Casting of Ductile Iron: A Numerical Approach’, AFS Transactions, 1987, 613-616. 2. Lerner V: ‘Mathematical Foundation of Information Macrodynamics’, International Journal ‘Systems Analysis-Modeling-Simulation’, Vol. 26, 1996, 119-184. 3. Lerner V: ‘ Informational Macrodynamics: System Analysis Modeling & Simulation Methodologies’, Proceedings of the 1998 Winter Simulation Conference, Washington, December 1998, 125-133. 4. Lerner V., Talyanker M: ’Informational Geometry for Biology and Environmental Applications’, Proceedings of the 2000 Western MultiConference, San Diego, January 2000,79-84. 5. Lebau C: ‘Properties of Continuous Cast Austempered Ductile Iron Bar,’ 2nd International Conference on Austempered Ductile Iron: Your means to improve Performance, Productivity and Cost, Rackham School, University of Michigan, Ann Arbor, MI, ASME, March 17-19, 1985, 215-226. 6. Lerner Y., Griffin G: “Developments in Continuous Casting of Gray and Ductile Iron’, Modern Casting, November 1997, pp.41-44. 7. Lerner Y: ‘Continuous Casting of Ductile Iron. Solidification, Microstructure, and Properties,’ 50th Electric Furnace Conference Proceedings, ISS, V. 50, Atlanta, USA, December 13-15, 1992, 331-340. 374