# Numerical modelling of steady detonations with the CREST reactive burn model

## Abstract

Watt et al. [J Eng Math 75(1):1–14, 2012] have shown that one can obtain good results for the propagation of detonation waves in cylindrical charges by assuming that the post-shock streamlines are straight. In this paper, we compare this Straight Streamline Approximation (SSA) to high-resolution Direct Numerical Simulations (DNS) for different models of explosives. We find that the SSA is less accurate for realistic explosion models than it is for polytropic equations of state with power-law reaction rates.

## Keywords

Detonation Diameter effects Explosives## 1 Introduction

### 1.1 Detonation of condensed phase explosives

A detonation wave is a combustion wave that propagates through explosive material at supersonic velocity. At the leading edge of a detonation wave, there is a shock that increases the pressure, density and temperature of the explosive sufficiently to initiate chemical reactions. The heat released by the reactions supports the propagation of the shock wave. These explosives are often heterogeneous in their composition, e.g. plastic bonded explosives, ammonium nitrate and fuel oil, etc., which can lead to complex microstructures with pores, cracks and voids. It is very difficult to model such explosives from first principles since the rate of heat release can be very sensitive to their mechanical structure [1].

Confinement of an explosive enables the detonation to occur at radii below the critical radius for an unconfined charge. This is because the confiner reduces the lateral expansion of the gaseous products, thus maintaining a higher pressure in the detonation reaction zone than in the unconfined case [2].

### 1.2 Modelling

One of the primary motivations for detonation modelling is the high cost of performing experiments. Such models can provide understanding as well as predictions of the behaviour of detonations in more complex geometries than a simple cylindrical charge. This means that the models can be calibrated with simple experiments and then applied to cases that are relevant to applications. This can reduce the need for expensive experiments.

Since detonation in heterogeneous explosives is complex, current models are phenomenological and attempt to capture the burning of the explosive at the continuum level. At this scale, the inviscid reactive Euler equations are assumed to provide an accurate description of the physics. These need to be supplemented by an equation of state for the explosive and its products together with equations describing the chemical reactions [3]. Equations of state are usually multiphase with separate equations for each phase. The phases are assumed to be in pressure (mechanical) equilibrium and moving with the same velocity. Popular choices of equations of state include Jones–Wilkins–Lee (JWL), Williamsburg, polytropic and finite-strain. Apart from the polytropic EOS, these equations of state are empirical and must be calibrated with experimental data. The equations governing the chemical reaction rate are also empirical and are typically calibrated with ignition timing and size-effect data [4, 5]. The reaction rates are generally related to the shock strength via a dependence upon one of the thermodynamic variables, such as pressure [6], temperature [7] and entropy [8].

The problem can be solved by numerical integration of the full time dependent Euler equations in a Direct Numerical Simulation (DNS) [9]. Although this produces an accurate solution provided the numerical resolution is high enough, it is computationally expensive. A number of simplified detonation models have therefore been developed, which have much lower computational cost. The first such model was the Wood–Kirkwood model [10], but this requires experimental data to calibrate the relationship between the shock front curvature and the charge radius [5]. An improvement to the Wood–Kirkwood model came with development of the Detonation Shock Dynamics (DSD) model [11, 12] which, at leading order, avoids the need for such calibration. It is based upon the assumption that the detonation velocity is close to the CJ velocity and that the shock front curvature is small. Although reasonably accurate for ideal explosives where the detonation velocity does not depart significantly from the CJ velocity, it tends to give too low detonation velocities for non-ideal explosives, where the detonation velocity can be as low as 50% of the CJ velocity [13].

Watt et al. [14] have developed a more sophisticated model, the Straight Streamline Approximation (SSA) model. This makes no assumptions about the shock front curvature, instead it assumes that the post-shock streamlines are straight. They showed that it can accurately reproduce the velocity–radius curves from high-resolution DNS calculations for non-ideal explosives. However, they only considered the simple case of a polytropic equation of state and power-law reaction rate. It is yet to be determined whether the SSA model produces accurate results for different equations of state and forms of the reaction rate. In this paper, we explore the accuracy of the SSA model with a variety of equations of state and expressions for the reaction rate.

## 2 Governing equations

### 2.1 Reactive Euler equations

*p*the pressure,

*e*the internal energy per unit mass and \(\lambda \) the chemical reaction progress variable (ranging from 0 for no reaction to 1 for complete reaction).

*W*is the chemical reaction rate, which the depends upon \(\lambda \) and local thermodynamic variables.

### 2.2 Explosive model

CREST is a model for condensed phase explosives with a two-phase equation of state and an entropy dependent reaction rate. The use of an entropy based reaction rate is unique to this model: it being more usual to use pressure or temperature [5, 6, 11, 15]. It has been shown that CREST can accurately reproduce run-time to detonation and radius effect curves [16]. DNS have been carried out using this model [8, 16, 17, 18], but it has not yet been combined with any steady state detonation models such as the Wood–Kirkwood model or DSD.

The two phases in the equation of state are an unreacted solid phase and a fully reacted gaseous phase. The equation of state for each phase can be written in the form \(e_k=e_k(p_k,\rho _k)\) where subscript *k* denotes the phase, *s* for the solid and *g* for the gas.

*A*,

*B*, \( R_1\), \(R_2 \), \(\rho _{0s}\), and

*w*are constants.

Equation of state and reaction rate parameters for the CREST reactive burn models EDC37 and PBX9502

Parameter | EDC37 | PBX9502 | Units |
---|---|---|---|

\(\rho _0\) | 1.8445 | 1.890 | \(\text {g cm}^{-3}\) |

\(\rho _{0s}\) | 1.8445 | 1.942 | \(\text {g cm}^{-3}\) |

| 6.642021 | 4.603 | \(\text {Mbar}\) |

| 0.2282927 | 0.09544 | \(\text {Mbar}\) |

\(R_1\) | 4.25 | 3.903 | |

\(R_2\) | 1.825 | 1.659 | |

\(\omega \) | 0.25 | 0.48 | |

\(E_d\) | 0.0719557 | 0.0373 | \(\text {Mbar}\,\text {cm}^{-3}\,\text {g}^{-1}\) |

\(K_{0s}\) | 0.1424525 | 0.090314021 | \(\text {Mbar}\) |

\(A_1\) | 2.417494 | 0.246257 | |

\(A_2\) | 2.208027 | 11.44221 | |

\(A_3\) | 0.0 | 0.0 | |

\(B_1\) | 0.0 | 16.8477 | |

\(B_2\) | 0.0 | 6.534913 | |

\(B_3\) | 0.0 | 0.0 | |

\(f_s\) | 0.0 | 0.05 | |

\(\delta \) | 0.0 | 0.021322 | |

\(\varGamma _1\) | 32.33557 | 126.4052 | |

\(\varGamma _2\) | 3.596933 | 6.554447 | |

\(\gamma _{00}\) | 0.4 | 0.4 | |

| 2.0 | 2.0 | |

\(c_0\) | \(2.0\times 10^8\) | \(2.0\times 10^7\) | \(\,\upmu \text {s}^{-2}\,(\text {Mbar}\,\text {cm}^3\text {g})^{-c_1}\) |

\(c_1\) | 2.0 | 2.5 | |

\(c_2\) | \(2.2\times 10^8\) | \(8.0\times 10^6\) | \(\,\upmu \text {s}^{-2}\,(\text {Mbar}\,\text {cm}^3\text {g})^{-c_3}\) |

\(c_3\) | 2.5 | 2.5 | |

\(c_6\) | 0.0 | \(1.8\times 10^{12}\) | \(\,\upmu \text {s}^{-1}\) |

\(c_8\) | \(1.6\times 10^{-4}\) | 0.0 | \(\,\upmu \text {s}^{-1}\,(\text {Mbar}\,\text {cm}^3\text {g}^{-1})^{c_9}\) |

\(c_9\) | 1.0 | 1.0 | |

\(c_{10}\) | \(4.0\times 10^5\) | \(3.0\times 10^{13}\) | \(\,\upmu \text {s}^{-1}\,(\text {Mbar}\,\text {cm}^3\text {g}^{-1})^{-c_{11}}\) |

\(c_{11}\) | 1.8 | 1.25 | |

\(c_{12}\) | 0.0 | 5.0 | |

\(c_{13}\) | 0.0 | 0.0012 | \(\text {Mbar}\,\text {cm}^3\text {g}^{-1}\) |

## 3 Streamline detonation model

Watt et al. [14] have developed an approximate model for detonations in cylindrical charges which make an assumption about the shape of the streamlines and then solves the equations along the streamlines. The simplest approximation is to assume that the post-shock streamlines are straight, the straight streamline approximation (SSA). This gave results which were in good agreement with DNS for both confined and unconfined charges for a polytropic EOS and a power-law reaction rate. In particular, they obtained much more accurate diameter effect curves and failure diameters for both ideal and non-ideal explosives than the standard DSD method. However, although it works well for a polytropic EOS and power law reaction rate it is not known whether this is true for other EOS and reaction rates.

### 3.1 Governing equations in streamline coordinates

*r*,

*z*) coordinates, this satisfies

*u*and

*v*are the velocities in the

*r*and

*z*directions, respectively. Equations (18) and (19) give

### 3.2 Master equation in *v*

*z*velocity,

*v*. Downstream of the shock there is no dissipation and the flow is isentropic if there is no reaction. We must therefore have

*Q*is the thermicity parameter [7]. To determine

*Q*in this case, we start with

*z*in the continuity equation (21)

*u*/

*v*using (20) gives

*z*derivative of the pressure, we use (22) to eliminate \(\partial p / \partial \psi \) in (23)

*u*from (20), we get

*z*velocity,

*v*

*v*and can be solved together with Eqs. (25), (33) and (34) for \(\rho (z), p(z), u(z), v(z)\) given the boundary conditions at the shock. However, since the streamline shape is not known, we must make some assumptions about their shape.

### 3.3 Straight streamline approximation

The simplest assumption is that the streamlines are straight as shown in Figure 2. In the frame of the shock, the upstream streamlines are parallel to the axis, and the velocity is \(-D_0\), where \(D_0\) is the detonation speed. At the point \((r_f,z_f)\) where the streamlines intersect the shock, they are deflected by the curved shock and are then assumed to remain straight until they reach the sonic locus.

*f*is given by the solution to the eigenvalue problem along the streamline. The solution of this equation with the boundary condition

The derivative \(\mathrm {d}{F}/\mathrm {d}{z_f^\prime }\) in (44) is found by taking the derivative of (39). With CREST the shock relations must be solved numerically, thus \(\mathrm {d}{F}/\mathrm {d}{z_f^\prime }\) must be computed numerically. A central difference approximation was used in this case.

### 3.4 Solving the system

To solve the system, we choose a detonation velocity \(D_0\) and then compute the shock surface by integrating equation (45) from the axis to the charge edge using the boundary conditions (46). For an unconfined charge, the charge edge is determined by the condition that the post-shock flow be sonic [11]. If the charge is confined, then the condition at the charge edge must be determined by a shock polar analysis. In either case, this gives \(D_0\) as a function of charge diameter.

As \(|z_f^\prime |\) increases, both the post-shock Mach number and streamline deflection increase monotonically until the streamline deflection reaches a maximum at the Crocco point [21]. For larger \(|z_f^\prime |\), the streamlines converge behind the shock, which means that \(\mathrm {d} F/\mathrm {d} \psi \) is negative and hence from (44) so is \(\partial ^2 r / \partial \psi \partial z\). The master equation, (35), has no solution in this case so that one cannot integrate to the sonic point when the Crocco point occurs before the sonic point. This is the case for the CREST EOS, so we simply assumed that the charge edge was at the Crocco point for unconfined charges. The impact on the accuracy of the results is discussed in Sect. 5.3.

### 3.5 DSD model

*v*is the velocity normal to the shock, \(D_n\) is the normal velocity of the shock and \(\kappa \) is its local total curvature. Using (26), we can derive the master equation for this model

This provides a relation for \(z_f^{\prime \prime }\) of the same form as (45), which can be integrated to give the shock shape. For a given detonation velocity, the charge radius is determined by integrating along the shock front from the charge centre out to the edge. As for the SSA, the Crocco point was used as the boundary condition at the charge edge.

A solution can only be obtained for \(D_n\) if \(\kappa \) is below a critical value, \(\kappa _c\), at which point \(D_n = D_c\). For \(\kappa < \kappa _c\) there are two solutions for \(D_n\), an upper stable branch with \(D_n > D_c\) and a lower unstable branch with \(D_n < D_c\) [23]. Since \(D_n\) decreases as we integrate out from the axis, it may happen that it becomes equal to \(D_c\) and we have to start using the value of \(\kappa \) from the unstable branch. Since this occurs near the failure radius, it is clear that this model is somewhat dubious for such cases.

## 4 Numerical method

In order to explore the accuracy of the SSA, high-resolution numerical calculations of the cylindrical charge problem were performed using Falle’s MG code [24]. This uses a second-order, conservative, upwind scheme to solve the full time-dependent Euler equations. The HLLC approximate Riemann solver [25] was used to compute the fluxes since it maintains stationary contacts and slip lines. This is important for 2D calculations in which there is a sharp interface between the explosive and the external medium. For reasons to be explained later, it is necessary to use a reasonably diffusive Riemann solver.

MG uses Adaptive Mesh Refinement (AMR) so that the grid is only refined where necessary. There are *N* grid levels, \(G^0,G^1,\ldots ,G^{N-1}\) with mesh spacings \(h_n = h_0/2^n\), where \(h_0\) is that on \(G^0\). \(G^0\) and \(G^1\) cover the whole domain, but finer grids need not do so. Refinement is on a cell-by-cell basis and is controlled by error estimates based on the difference between solutions on different grids i.e. the difference between the solutions on \(G^{n-1}\) and \(G^n\) determines refinement to \(G^{n+1}\). An additional refinement condition, based upon the chemical reaction rate, was included to ensure that the solution was always resolved to the finest level in the reaction zone.

Like all shock-capturing codes, MG works with the conserved variables, \((e,\rho ,\lambda )\), but the pressure and densities of the phases are needed to compute the fluxes and source terms. These were obtained from the pressure equilibrium condition (7) and mixture equations, (5), (6) using a secant method iterating on the solid density. To prevent reaction ahead of the shock front, the reaction source term was set to zero for \(p < 0.05\,\text {Mbar}\). For \(\lambda < 0.02\), the solid-phase equation of state was used; otherwise, the mixture model was used.

### 4.1 One-dimensional calculations

Comparison between the detonation velocity obtained from the steady-state ZND model and DNS for two CREST explosive models

Explosive | ZND/\(\text {cm}\,\,\upmu \text {s}^{-1}\) | DNS/\(\text {cm}\,\,\upmu \text {s}^{-1}\) |
---|---|---|

EDC37 | 0.87693 | 0.87695 |

PBX9502 | 0.75974 | 0.75970 |

### 4.2 Two-dimensional calculations

The numerical domain was \(0 \le r \le 2\), \(0 \le z \le z_u\) in cylindrical polar (*r*, *z*) coordinates with the explosive in \(0< r < 1\) and an inert, low density, confiner in \(1<r <2\). \(z_u\) was such that the shock was in the interior of the domain and therefore depended on the particular case. For the unconfined cases, the confiner density was \(\rho _c = 0.9\,\text {g}\,\text {cm}^{-3}\) for EDC37 and \(\rho _c = 1.5\,\text {g}\,\text {cm}^{-3}\) for PBX9502, which are small enough for the expansion fan at the charge edge to be transonic. The boundary conditions were symmetric on \(r = 0\) and zero gradient on the other boundaries.

The radius of the explosive in physical units was varied by multiplying the rate constant in the source term by an appropriate factor. The detonation was initiated by imposing a high pressure region in \(0 \le r \le 1\), \(0 \le z \le 2\) and the detonation was then allowed to propagate until it became steady. The *z* coordinate of the shock front on the axis was tracked as a function of time and the resulting \(z-t\) data used to compute the detonation velocity.

### 4.3 Effect of numerical resolution

Table 3 shows the results for EDC37 for different numerical resolutions. It can be seen that 320 cells/\(\text {mm}\) are required to obtain well-converged results for solid-phase entropy and the detonation velocity. This corresponds to 69 cells in the reaction zone which is consistent with previous work in which it was found that \(\simeq 50\) cells were required for convergence [13].

Properties of axial solution as a function of numerical resolution (cells/\(\text {mm}\)) for EDC37 with charge radius \(R=0.67\,\text {cm}\)

Resolution | Entropy/\(10^{-2}\,\text {Mbar}\,\text {g}\,\text {cm}^{-3}\) | \(\lambda _{\text {max}}\) |
| CIRZ | \(\lambda _{\text {shk}}\) |
---|---|---|---|---|---|

20 | 0.85 | 0.856 | 0.768 | 6 | 0.65 |

40 | 1.23 | 0.907 | 0.810 | 9 | 0.40 |

80 | 1.33 | 0.928 | 0.809 | 17 | 0.05 |

160 | 1.27 | 0.912 | 0.801 | 34 | 0.005 |

320 | 1.25 | 0.911 | 0.798 | 69 | 0.002 |

640 | 1.24 | 0.910 | 0.797 | 140 | 0.001 |

## 5 Results and discussion

### 5.1 Unconfined calculations

#### 5.1.1 CREST reaction rate

#### 5.1.2 Power-law reaction rate

The main difference between this reaction rate and that of the CREST model is that the reaction rate is a maximum at the shock and monotonically decreases, whereas for CREST the reaction has an induction zone immediately behind the shock before reaching its maximum value a significant distance behind the shock.

### 5.2 Confined calculation

Calculations were also performed with the CREST EDC37 model with confinement. Confinement reduces shock curvature and also prevents the flow from being sonic at the charge edge. This enables detonation waves to propagate in charges with a much smaller radius than in the unconfined case [2]. The reduced curvature means that the streamlines are straighter and the SSA should be more accurate than in unconfined detonations.

### 5.3 Failure to integrate to the charge edge

In Sect. 3.4, we showed that it is not possible to integrate to the charge edge for unconfined charges when the Crocco point occurs before the sonic point. Figure 12 shows how the post-shock Mach number varies with streamline deflection for EDC37, ANFO and polytropic equations of state. The ANFO equation of state was a pseudo-polytropic equation of state in which gamma is a quadratic function of the density [15]. It is clear that the sonic point occurs after the Crocco point both for ANFO and the solid-phase equation used by CREST, whereas they coincide for a polytropic EOS [28]. Figure 12 suggests that for detonation modelling of unconfined explosives with the SSA the Crocco point will define the boundary condition at the edge of the rate stick because for any realistic EOS the Crocco point is at a shock angle smaller than the angle required for sonic flow post shock.

Figure 13a, b shows the shock slope as a function of radius for \(D=0.861\,\text {cm}\,\,\upmu \text {s}^{-1}\) (near CJ) and \(D=0.760\,\text {cm}\,\,\upmu \text {s}^{-1}\) (near failure) for EDC37 with the CREST reaction rate. Both cases show the same general trend: the SSA and DNS are in good agreement for \(r<R/2\) and thereafter the shock slope for SSA increases more rapidly than for DNS. Inspection of Fig. 6 in conjunction with Fig. 13b shows that the radius at which the SSA and DNS begin to diverge closely matches the radius at which the peak reaction rate reduces significantly compared to the rate on axis. As the reaction rate is smaller here, then curvature effects will more significantly affect the flow. The DNS shows that the curvature is large near the charge edge, which means that the Crocco point occurs at almost the same radius as the sonic point. The discrepancy between SSA and DNS in Fig. 4 must therefore be due to streamline curvature, rather than our inability to integrate past the Crocco point.

Figure 14a, b is similar to Fig. 13a, b except that the reaction rate is a power-law one. For a detonation speed near to CJ (Fig. 14a), the SSA and DNS are in reasonable agreement, but for the lower detonation speed shown in Fig. 14b, the SSA shock slope decreases more rapidly than the DNS. This suggests that for a power-law reaction rate, curvature effects only become significant at lower detonation velocities, which is why the agreement between the SSA and DNS diameter effect curves in Fig. 8 is worse at lower detonation velocities. The main difference between Figs. 14a, b and 13a, b is that the shock curvature diverges between the SSA and DNS at a radius much closer to the charge edge with the power-law reaction compared to the standard CREST reaction model. This suggests that for modelling with CREST curvature effects are significant at all detonation velocities and in a greater portion of the reaction zone than for a power-law reaction.

## 6 Streamline curvature

*s*is the shock slope. From [11], we have \(s^2 = (\gamma - 1)/ (\gamma + 1) = 1 / 2\) at the sonic point, so \(s^2 < 1 /2\). Note that with \(W = 0\), this agrees with the expression given in [21] which does not include reactions.

The first term in the square braces is proportional to the reaction rate and is strictly positive. The second term, which is proportional to the shockwave curvature \(\kappa \), is of the opposite sign since \(\kappa < 0\) and \(s^2 < 1 /2\). This is also true for the CREST solid-phase equation of state, although in this case the post-shock streamline curvature must be found numerically from Eqs. (68) and (55) in Appendix A. In either case, the effect of the reaction term is to reduce the streamline curvature.

*m*and

*n*are constants [14].

## 7 Conclusions

We have compared the predictions of the SSA with DNS for the CREST model and have shown that it gives accurate values of the detonation velocity for unconfined cylindrical charges whose diameter is significantly larger than the failure diameter. However, for smaller diameters, it performs poorly compared with previous results for polytropic equations of state with a power law reaction rate [14].

There are two reasons for this. The first is that the CREST reaction rate is zero at the shock, whereas it is a maximum there for a power-law reaction rate. Since reaction at the shock reduces the curvature of the streamlines for a given shock curvature, this means that the assumption that the streamlines are straight with slope equal to that at the shock is less well satisfied for CREST. Secondly, for the CREST equation of state, the sonic point on the shock polar occurs after the Crocco point at which the streamline deflection is a maximum. Since the SSA requires the streamline deflection to increase with radius, this makes it impossible to integrate to the charge edge for an unconfined charge. The DNS calculations show that the Crocco point occurs very close to the charge edge, which means that the main source of inaccuracy is the streamline curvature at the shock, not the equation of state. We have confirmed this by showing that SSA gives excellent results for the CREST equation of state with a power-law reaction rate. Since such models are widely used, there is still a place for the SSA in explosives modelling.

The results are much better for confined charges since the shock curvature is smaller and there is no difficulty in integrating to the charge edge in this case. Although the properties of explosives are usually determined from experiments with unconfined charges, the confined case is the most relevant to applications. The SSA approximation is therefore still usable for the CREST explosive models and others with similar properties.

The only way to improve the accuracy of SSA is add the effect of curvature. A possible approach is to introduce some curvature of the streamlines that depends upon a set of adjustable parameters. The complete solution can be constructed for some initial guess for the values of these parameters on each streamline. The extent to which this solution fails to satisfy the equations can then be used to construct a global error. Minimization of this error determines the optimal values of the parameters on each streamline. This Variational Streamline Approach (VSA) is clearly more expensive than SSA, but can be made significantly more accurate, particularly if there are several parameters on each streamline. Preliminary calculations with a single parameter determining the curvature near the shock show considerable promise.

Finally, we have been entirely concerned with the verification of the SSA with CREST by comparison with DNS, not the validation of the CREST model by its ability to predict the results of experiments. As with any model, CREST must be properly calibrated to other experimental data, and the SSA can be useful for this. Even for cases in which it is not very accurate, it can provide initial guesses for the parameters in explosion models which can then be refined with DNS calculations.

## Notes

### Acknowledgements

The authors would like to acknowledge funding provided by EPSRC and AWE.

## References

- 1.Tarver CM, Chidester SK, Nichols AL (1996) Critical conditions for impact-and shock-induced hot spots in solid explosives. J Phys Chem 100(14):5794–5799CrossRefGoogle Scholar
- 2.Sharpe G, Bdzil J (2006) Interactions of inert confiners with explosives. J Eng Math 54(3):273–298MathSciNetCrossRefzbMATHGoogle Scholar
- 3.Zhang F (2009) Shock wave science and technology reference library. Springer, BerlinCrossRefGoogle Scholar
- 4.Vorthman J, Andrews G, Wackerle J (1985) Reaction rates from electromagnetic gauge data. Technical Report, Los Alamos National Laboratory, New MexicoGoogle Scholar
- 5.Schoch S, Nikiforakis N, Lee BJ, Saurel R (2013) Multi-phase simulation of ammonium nitrate emulsion detonations. Combust Flame 160(9):1883–1899CrossRefGoogle Scholar
- 6.Lee E, Tarver C (1980) Phenomenological model of shock initiation in heterogeneous explosives. Phys Fluids 23:2362–2372CrossRefGoogle Scholar
- 7.Fickett W, Davis W (1979) Detonation: theory and experiment. Courier Dover Publications, MineolaGoogle Scholar
- 8.Handley C, Whitworth N, James H, Lambourn B, Maheswaran M (2010) The CREST reactive-burn model for explosives. In: EPJ web of conferences, vol 10, 4. EDP SciencesGoogle Scholar
- 9.Sharpe GJ, Falle S (2000) One-dimensional nonlinear stability of pathological detonations. J Fluid Mech 414:339–366MathSciNetCrossRefzbMATHGoogle Scholar
- 10.Wood WW, Kirkwood JG (1954) Diameter effect in condensed explosives. The relation between velocity and radius of curvature of the detonation wave. J Chem Phys 22:1920–1924CrossRefGoogle Scholar
- 11.Bdzil J (1981) Steady-state two-dimensional detonation. J Fluid Mech 108:195–226CrossRefzbMATHGoogle Scholar
- 12.Yao J, Stewart D (1996) On the dynamics of multi-dimensional detonation. J Fluid Mech 309:225–275MathSciNetCrossRefzbMATHGoogle Scholar
- 13.Aslam T, Bdzil J, Hill L (1998) Extensions to DSD theory: Analysis of PBX9502 rate stick data. In: 11th International symposium on detonation, pp 21–29Google Scholar
- 14.Watt SD, Sharpe GJ, Falle SA, Braithwaite M (2012) A streamline approach to two-dimensional steady non-ideal detonation: the straight streamline approximation. J Eng Math 75(1):1–14CrossRefzbMATHGoogle Scholar
- 15.Sharpe G, Braithwaite M (2005) Steady non-ideal detonations in cylindrical sticks of explosives. J Eng Math 53(1):39–58CrossRefzbMATHGoogle Scholar
- 16.Whitworth N (2008) Mathematical and numerical modelling of shock initiation in heterogeneous solid explosives. PhD thesis, Cranfield UniversityGoogle Scholar
- 17.Whitworth N (2007) Application of the CREST reactive burn model to two-dimensional PBX 9502 explosive experiments. In: Shock compression of condensed matter—2007: proceedings of the conference of the American Physical Society topical group on shock compression of condensed matter, vol 955. AIP Publishing, Melville, pp 881–884Google Scholar
- 18.Whitworth N, Childs M (2015) Determination of detonation wave boundary angles via direct numerical simulations using CREST. In: APS shock compression of condensed matter meeting abstracts 1:2002Google Scholar
- 19.Zukas JA, Walters W, Walters WP (2002) Explosive effects and applications. Springer, New YorkGoogle Scholar
- 20.Cowperthwaite M (1981) A constitutive model for calculating chemical energy release rates from the flow fields in shocked explosives. In: Seventh international symposium on detonation, pp 498–505Google Scholar
- 21.Rao P (1973) Streamline curvature and velocity gradient behind curved shocks. AIAA J 11(9):1352–1354CrossRefGoogle Scholar
- 22.Hill L, Bdzil JB, Aslam TD (1998) Front curvature rate stick measurements and detonation shock dynamics calibration for PBX 9502 over a wide temperature range. In: Eleventh international detonation symposium, pp 1029–1037Google Scholar
- 23.Stewart DS (1998) The shock dynamics of multidimensional condensed and gas-phase detonations. In: International symposium on combustion, vol 27(2), pp 2189–2205Google Scholar
- 24.Falle S (1991) Self-similar jets. Mon Not R Astron Soc 250(3):581–596CrossRefGoogle Scholar
- 25.Toro EF, Spruce M, Speares W (1994) Restoration of the contact surface in the HLL-Riemann solver. Shock Waves 4(1):25–34CrossRefzbMATHGoogle Scholar
- 26.Falle S, Komissarov S, Joarder P (1998) A multidimensional upwind scheme for magnetohydrodynamics. Mon Not R Astron Soc 297(1):265–277CrossRefGoogle Scholar
- 27.Gambino JR, Tarver CM, Springer HK (2018) Numerical parameter optimizations of the ignition and growth model for a HMX plastic bonded explosive. J Appl Phys 124:195901-1–195901-6CrossRefGoogle Scholar
- 28.Cartwright M (2016) Modelling of non-ideal steady detonations. PhD thesis, University of LeedsGoogle Scholar
- 29.Thomas T (1947) On curved shock waves. J Maths Phys Camb 26(1):62–68MathSciNetCrossRefzbMATHGoogle Scholar
- 30.Mölder S (2016) Curved shock theory. Shock Waves 26(4):337–353CrossRefGoogle Scholar

## Copyright information

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.