ANFIS Modeling for Prediction of Particle Size in Nozzle Assisted Solvent-Antisolvent Process for Making Ultrafine CL-20 Explosiv

Physical properties such as particle size, surface area and shape of explosive control the rapidity and reliability of initiation, and detonation and thus determine the performance of an explosive device such as slapper detonators. In this paper, Nozzle assisted solvent/antisolvent (NASAS) process for recrystallisation of CL-20 explosive is established. Many process parameters are involved which affect the particle size of the explosive. Therefore an accurate prediction of particle size is required to tailor the particle size. In the present work, an intelligent algorithm is applied to build a simplified relationship between recrystallization process parameters and particle size. This can be used to predict explosive particle size with a wide range of process parameters through an adaptive neuro-fuzzy inference system (ANFIS). The model is trained using experimental data obtained from design of experiment techniques utilizing a MATLAB software. Six process parameters such as Solution pressure, Antisolvent pressure, Antisolvent temperature, Stirrer speed, Solution concentration and Nozzle diameter are used as input variables of the model and the particle size is used as the output variable. The predicted results are in close agreement with experimental values and the accuracy of the model has been tested by comparing the simulated data with actual data from the explosive recrystallization experiments and found to be inacceptable range with maximum absolute percentage error of 11.52 %. The ultrafine CL-20 prepared by NASAS process is used in Slapper detonator application. The threshold initiation voltages for CL-20 based slapper detonator is found to be in the range of 0.9 kV with standard deviation of ±0.1 kV. Introduction The physical properties such as crystal particle size, shape, morphology, crystalline imperfections, purity and microstructure of the inter-crystalline voids of an existing explosive can be altered. There are wide variety of processes available for tailoring particle size and morphology of energetic materials such as solvent/non-solvent recrystallization[1],continuous crystallization of submicrometer energetic materials [2], spray flash evaporation [3]Yang et al. [4] obtained nanoTATB by using solvent/anti-solvent method with a particle size of 60 nm approximately through atomization of solution by a nozzle to small droplets and colliding rapidly with non-solvent flow. There is a need of mathematical model to predict particle characteristics as a function of process parameters to provide a basis for a computer based process control system. Shallu Gupta et al.[5,6], used micro nozzle assisted spraying process (MNASP) for recrystallization of Submicrometer Hexanitrostilbene (sm-HNS) Explosive. The process attributes were optimized using weighted average techniques of Analytical Network Process (ANP). The advantages of neural network based Explosion Shock Waves and High Strain Rate Phenomena Materials Research Forum LLC Materials Research Proceedings 13 (2019) 121-127 https://doi.org/10.21741/9781644900338-21 122 techniques include extreme computation, powerful memory and rapid learning from experimental data. Furthermore, it can predict an output parameter with accuracy even if the input parameter interactions are not completely understood[7, 8]. Artificial neural network (ANN) and multilayer perceptron (MLP) is widely established inartificial intelligence (AI) research where a nonlinear mapping between input and output parameters is required for a function approximation[9, 10]. Pannier et. al, have explained the application and general features of Fuzzy logic (FL)modeling, fuzzy sets, membership functions, and fuzzy clustering[11]. theoretical details of the neuro-fuzzy modeling can be found in [12, 13]. Moreover, however, relevant features and context that refer to the adopted means of neuro-fuzzy modeling, i.e., ANFIS [14] It is seen from literature that in spite of being powerful modeling tool, ANFIS has not been used in the study of explosive recrystallization process. A neuro-fuzzy technique called adaptive network based fuzzy inference system (ANFIS) combines fuzzy systems with neural networks, utilizing the learning characteristics of neural network and decision making capability of fuzzy systems. In this research work, application of ANFIS model is adopted for predicting the particle size of CL-20 explosive in solvent-antisolvent recrystallization process. Experimental Work The explosive material used in this research work was raw ε-CL-20 with a particle size in the range of 50 to 60 μm. In this research work, for making UF-CL20, a Nozzle Assisted Solvent-Antisolvent (NASAS) process has been designed, developed, fabricated and installed, as per schematic diagram shown in Fig. 2. The NASAS process was used to carry out 49 experiments for making UF-CL20 explosive. Based on design of experiments, six input parameters were considered which affect the output of the re-crystallized explosive i.e. particle size. The input parameters are solution pressure, anti-solvent pressure, anti-solvent temperature, stirrer speed, solution concentration and nozzle diameter. The output parameter i.e. particle size was used as the response variable. The UF-CL20 obtained by NASAS process was characterized as explained in the following section. Figure 1. Schematic of NASAS process Characterization The distribution of particle size for some of the samples under similar condition is shown in Fig. 3 with mean particle size of UF-CL20 as 2.61 μm with standard deviation of 0.242 μm. Total 42 Nos. of experiments were carried out to record the 42 data of input-output pairs of variables shown in Table 1 for ANFIS model. Recrystallised ultrafine CL-20 was characterized using XRD analysisto ensure crystalline nature XRD pattern showed the peaks at similar difraction angle as those of CL20 which exhibits a unique non-overlapping diffraction peak at 19.98 2θ, as shown in Fig.4. FTIR analysis was carried out to ensure there is no change in molecular structure after processing as shown in Fig.5. SEM photography showed the reduction of particle size and the morphology was Explosion Shock Waves and High Strain Rate Phenomena Materials Research Forum LLC Materials Research Proceedings 13 (2019) 121-127 https://doi.org/10.21741/9781644900338-21 123 also affected by the process parameters as shown in Fig. 6. The shape is a mix of polyhedral and nearly spherical geometry. The surface seems to be smooth with negligible defects/ cracks. Figure 2. Particle size distribution Figure 3. XRD pattern of processed CL20 Figure 4. FTIR Analysis Figure 5. SEM microphotograph Table 1. Experimental data of particle size Run Order Experime nt Solutio n Pressur e (bar) Antisolve nt Pressure (bar) Antisolvent Temperature (°C) Stirrer Speed (RPM) Solution Concentrati on (%) Nozzle Diameter (mm) Particl e Size (μm) 1 N-19 6 6 -9 800 5 0.7 5.63 2 N-20 7 7 -9 800 5 0.7 5.77 3 N-26A 5 1 3